Information Management and Applications of Intelligent ...

Mathematical Problems in Engineering

Information Management and Applications of Intelligent Transportation System

Guest Editors Chi-Chun Lo Kuo-Ming Chao Hsu-Yang Kung Chi-Hua Chen and Maiga Chang

Information Management and Applications of

Intelligent Transportation System




Guest Editors Chi-ChunLoKuo-MingChaoHsu-YangKung

Chi-Hua Chen and Maiga Chang

Copyright copy 2015 Hindawi Publishing Corporation All rights reserved

is is a special issue published in ldquoMathematical Problems in Engineeringrdquo All articles are open access articles distributed under theCreative Commons Attribution License which permits unrestricted use distribution and reproduction in any medium provided theoriginal work is properly cited

Editorial Board

MAbd El Aziz EgyptF Abed-Meraim FranceSilvia Abrahatildeo SpainPaolo Addesso ItalyClaudia Adduce ItalyRamesh Agarwal USAJuan C Aguumlero AustraliaR Aguilar-Loacutepez MexicoTarek Ahmed-Ali FranceHamid Akbarzadeh CanadaM N Akram NorwayMohammad-Reza Alam USAS Alfonzetti ItalyF Alhama SpainJuan A Almendral SpainLionel Amodeo FranceIgor Andrianov GermanySebastian Anita RomaniaRenata Archetti ItalyFelice Arena ItalySabri Arik TurkeyFumihiro Ashida JapanHassan Askari CanadaMohsen A Zaeem USAF Aymerich ItalySeungik Baek USAKhaled Bahlali FranceLaurent Bako FranceStefan Balint RomaniaAlfonso Banos SpainRoberto Baratti ItalyMartino Bardi ItalyA Beghdadi FranceA-H Bendada CanadaIvano Benedetti ItalyElena Benvenuti ItalyJamal Berakdar GermanyE Berjano SpainJean-Charles Beugnot FranceSimone Bianco ItalyDavid Bigaud FranceJonathan N Blakely USAPaul Bogdan USADaniela Boso ItalyA-O Boudraa France

F Braghin ItalyMichael J Brennan UKMaurizio Brocchini ItalyJulien Bruchon FranceJavier Bulduacute SpainTito Busani USAP Cacciola UKS Caddemi ItalyJose E Capilla SpainAna Carpio SpainMiguel E Cerrolaza SpainM Chadli FranceGregory Chagnon FranceChing-Ter Chang TaiwanMichael J Chappell UKKacem Chehdi FranceChunlin Chen ChinaXinkai Chen JapanFrancisco Chicano SpainHung-Yuan Chung TaiwanJoaquim Ciurana SpainJohn D Clayton USACarlo Cosentino ItalyPaolo Crippa ItalyErik Cuevas MexicoPeter Dabnichki AustraliaLuca DrsquoAcierno ItalyWeizhong Dai USAP Damodaran USAF Daneshmand CanadaFabio De Angelis ItalyS de Miranda ItalyF de Monte ItalyXavier Delorme FranceLuca Deseri USAY Dimakopoulos GreeceZhengtao Ding UKRalph B Dinwiddie USAMohamed Djemai FranceAlexandre B Dolgui FranceG S Dulikravich USABogdan Dumitrescu FinlandHorst Ecker AustriaAhmed El Hajjaji FranceFouad Erchiqui Canada

Anders Eriksson SwedenGiovanni Falsone ItalyHua Fan ChinaYann Favennec FranceG Fedele ItalyRoberto Fedele ItalyJacques Ferland CanadaJose R Fernandez SpainSimme Douwe Flapper Netherlandsierry Floquet FranceEric Florentin FranceFrancesco Franco ItalyTomonari Furukawa USAMohamed Gadala CanadaMatteo Gaeta ItalyZoran Gajic USACiprian G Gal USAUgo Galvanetto ItalyAkemi Gaacutelvez SpainRita Gamberini ItalyMaria Gandarias SpainArman Ganji CanadaXin-Lin Gao USAZhong-Ke Gao ChinaGiovanni Garcea ItalyFernando Garciacutea SpainLaura Gardini ItalyA Gasparetto ItalyV Gattulli ItalyOleg V Gendelman IsraelMergen H Ghayesh AustraliaAnna M Gil-Lafuente SpainHector Goacutemez SpainRama S R Gorla USAOded Gottlieb IsraelAntoine Grall FranceJason Gu CanadaQuang Phuc Ha AustraliaOfer Hadar IsraelMasoud Hajarian IranFreacutedeacuteric Hamelin FranceZhen-Lai Han Chinaomas Hanne SwitzerlandTakashi Hasuike JapanXiao-Qiao He China

MI Herreros SpainVincent Hilaire FranceEckhard Hitzer JapanJaromir Horacek Czech RepublicMuneo Hori JapanAndraacutes Horvaacuteth ItalyGordon Huang CanadaSajid Hussain CanadaAsier Ibeas SpainGiacomo Innocenti ItalyEmilio Insfran SpainNazrul Islam USAPayman Jalali FinlandReza Jazar AustraliaKhalide Jbilou FranceLinni Jian ChinaBin Jiang ChinaZhongping Jiang USANingde Jin ChinaGrand R Joldes AustraliaJoaquim Joao Judice PortugalT Kaczorek PolandTamas Kalmar-Nagy HungaryT Kapitaniak PolandHaranath Kar IndiaK Karamanos BelgiumC M Khalique South AfricaDo Wan Kim KoreaNam-Il Kim KoreaOleg Kirillov GermanyManfred Krafczyk GermanyFrederic Kratz FranceJurgen Kurths GermanyK Kyamakya AustriaDavide La Torre ItalyRisto Lahdelma FinlandHak-Keung Lam UKAntonino Laudani ItalyAimersquo Lay-Ekuakille ItalyMarek Lek PolandYaguo Lei Chinaibault Lemaire FranceStefano Lenci ItalyRoman Lewandowski PolandQing Q Liang AustraliaPanos Liatsis UKPeide Liu ChinaPeter Liu Taiwan

Wanquan Liu AustraliaYan-Jun Liu ChinaJean J Loiseau FrancePaolo Lonetti ItalyLuis M Loacutepez-Ochoa SpainVassilios C Loukopoulos GreeceV Lychagin NorwayFazal M Mahomed South AfricaYassir T Makkawi UKNoureddine Manamanni FranceDidier Maquin FranceP M Mariano ItalyBenoit Marx FranceGeampaposrard A Maugin FranceDriss Mehdi FranceRoderick Melnik CanadaPasquale Memmolo ItalyXiangyu Meng CanadaJose Merodio SpainLuciano Mescia ItalyLaurent Mevel FranceYuri V Mikhlin UkraineAki Mikkola FinlandHiroyuki Mino JapanPablo Mira SpainVito Mocella ItalyRoberto Montanini ItalyGisele Mophou FranceRafael Morales SpainAziz Moukrim FranceEmiliano Mucchi ItalyDomenico Mundo ItalyJose J Muntildeoz SpainGiuseppe Muscolino ItalyMarco Mussetta ItalyHakim Naceur FranceHassane Naji FranceDong Ngoduy UKTatsushi Nishi JapanBen T Nohara JapanMohammed Nouari FranceMustapha Nourelfath CanadaSotiris K Ntouyas GreeceRoger Ohayon FranceMitsuhiro Okayasu JapanEva Onaindia SpainJavier Ortega-Garcia SpainA Ortega-Montildeux Spain

Naohisa Otsuka JapanErika Ottaviano ItalyA Paipetis GreeceA Palmeri UKAnna Pandol ItalyElena Panteley FranceManuel Pastor SpainPubudu N Pathirana AustraliaFrancesco Pellicano ItalyHaipeng Peng ChinaMingshu Peng ChinaZhike Peng ChinaMarzio Pennisi ItalyMatjaz Perc SloveniaFrancesco Pesavento ItalyMaria do Rosaacuterio Pinho PortugalAntonina Pirrotta ItalyVicent Pla SpainJavier Plaza SpainJean-Christophe Ponsart FranceMauro Pontani ItalyStanislav Potapenko CanadaSergio Preidikman USAChristopher Pretty New ZealandCarsten Proppe GermanyLuca Pugi ItalyYuming Qin ChinaDane Quinn USAJose Ragot FranceKumbakonam Ramamani Rajagopal USAGianluca Ranzi AustraliaSivaguru Ravindran USAAlessandro Reali ItalyOscar Reinoso SpainNidhal Rezg FranceRicardo Riaza SpainGerasimos Rigatos GreeceJoseacute Rodellar SpainRosana Rodriguez-Lopez SpainIgnacio Rojas SpainCarla Roque PortugalAline Roumy FranceDebasish Roy IndiaRubeacuten Ruiz Garciacutea SpainAntonio Ruiz-Cortes SpainIvan D Rukhlenko AustraliaMazen Saad FranceKishin Sadarangani Spain

Mehrdad Saif CanadaMiguel A Salido SpainRoque J Saltareacuten SpainFrancisco J Salvador SpainAlessandro Salvini ItalyMaura Sandri ItalyMiguel A F Sanjuan SpainJuan F San-Juan SpainRoberta Santoro ItalyIlmar Ferreira Santos DenmarkJoseacute A Sanz-Herrera SpainNickolas S Sapidis GreeceEvangelos J Sapountzakis GreeceAndrey V Savkin AustraliaValery Sbitnev Russiaomas Schuster GermanyMohammed Seaid UKLot Senhadji FranceJoan Serra-Sagrista SpainLeonid Shaikhet UkraineHassan M Shanechi USASanjay K Sharma IndiaBo Shen GermanyBabak Shotorban USAZhan Shu UKDan Simon USALuciano Simoni ItalyChristos H Skiadas GreeceMichael Small AustraliaFrancesco Soldovieri ItalyRaaele Solimene Italy

Ruben Specogna ItalySri Sridharan USAIvanka Stamova USAYakov Strelniker IsraelSergey A Suslov Australiaomas Svensson SwedenAndrzej Swierniak PolandYang Tang GermanySergio Teggi ItalyAlexander Timokha NorwayRafael Toledo SpainGisella Tomasini ItalyFrancesco Tornabene ItalyAntonio Tornambe ItalyFernando Torres SpainFabio Tramontana ItalySeacutebastien Tremblay CanadaIrina N Trendalova UKGeorge Tsiatas GreeceAntonios Tsourdos UKVladimir Turetsky IsraelMustafa Tutar SpainEfstratios Tzirtzilakis GreeceFilippo Ubertini ItalyFrancesco Ubertini ItalyHassan Ugail UKGiuseppe Vairo ItalyKuppalapalle Vajravelu USARobertt A Valente PortugalPandian Vasant MalaysiaMiguel E Vaacutezquez-Meacutendez Spain

Josep Vehi SpainKalyana C Veluvolu KoreaFons J Verbeek NetherlandsFranck J Vernerey USAGeorgios Veronis USAAnna Vila SpainRafael J Villanueva SpainUchechukwu E Vincent UKMirko Viroli ItalyMichael Vynnycky SwedenJunwu Wang ChinaShuming Wang SingaporeYan-WuWang ChinaYongqi Wang GermanyDesheng D Wu CanadaYuqiang Wu ChinaGuangming Xie ChinaXuejun Xie ChinaGen Qi Xu ChinaHang Xu ChinaXinggang Yan UKLuis J Yebra SpainPeng-Yeng Yin TaiwanIbrahim Zeid USAHuaguang Zhang ChinaQingling Zhang ChinaJian Guo Zhou UKQuanxin Zhu ChinaMustapha Zidi FranceAlessandro Zona Italy

Contents

Information Management and Applications of Intelligent Transportation System Chi-Chun LoKuo-Ming Chao Hsu-Yang Kung Chi-Hua Chen and Maiga ChangVolume 2015 Article ID 613940 2 pages

Novel Encoding and Routing Balance Insertion Based Particle SwarmOptimization with Application to

Optimal CVRP Depot Location Determination Ruey-Maw Chen and Yin-Mou ShenVolume 2015 Article ID 743507 11 pages

AMethod for Driving Route Predictions Based on Hidden MarkovModel Ning Ye Zhong-qin WangReza Malekian Qiaomin Lin and Ru-chuan WangVolume 2015 Article ID 824532 12 pages

Detecting Trac Anomalies in Urban Areas Using Taxi GPS Data Weiming Kuang Shi Anand Huifu JiangVolume 2015 Article ID 809582 13 pages

Identifying Key Factors for Introducing GPS-Based Fleet Management Systems to the Logistics

Industry Yi-Chung Hu Yu-Jing Chiu Chung-Sheng Hsu and Yu-Ying ChangVolume 2015 Article ID 413203 14 pages

Image-Based Pothole Detection System for ITS Service and RoadManagement System Seung-Ki RyuTaehyeong Kim and Young-Ro KimVolume 2015 Article ID 968361 10 pages

EditorialInformation Management and Applications ofIntelligent Transportation System

Chi-Chun Lo1 Kuo-Ming Chao2 Hsu-Yang Kung3 Chi-Hua Chen145 and Maiga Chang6

1Department of Information Management and Finance National Chiao Tung University 1001 University Road Hsinchu 300 Taiwan2Department of Computing Coventry University Priory Street Coventry CV1 5FB UK3Department of Management Information Systems National Pingtung University of Science and Technology1 Shuefu Road Neipu Pingtung 912 Taiwan4Telecommunication Laboratories Chunghwa Telecom Co Ltd 99 Dianyan Road Yangmei District Taoyuan 326 Taiwan5Department of Communication and Technology National Chiao Tung University 1001 University Road Hsinchu 300 Taiwan6School of Computing and Information Systems Athabasca University 1 University Drive Athabasca AB Canada T9S 3A3

Correspondence should be addressed to Chi-Hua Chen chihua0826gmailcom

Received 5 August 2015 Accepted 11 August 2015

Copyright copy 2015 Chi-Chun Lo et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

1 Introduction

The rise of economic growth and technology advance hasled to increasing demand of the intelligent transportationsystem (ITS) for traffic service How to construct real-timeinformation systems of ITS has become more important[1] Real-time traffic information such as average vehiclespeed travel time traffic flow and traffic congestion canbe used by road users and the ministry of transportationto improve the level of service for road ways Severalapproaches have been developed to collect and send real-time traffic information to traffic information centre viavarious networks (eg vehicular ad hoc network (VANET)[2] universal mobile telecommunications system (UMTS)[3] and long-term evolution (LTE) [4]) vehicle detector [5]global position system- (GPS-) based probe car reporting[6] cellular floating vehicle data (CFVD) [7] and so forthFurthermore information and communications technology(ICT) can be used to analyse the real-time traffic informationto forecast the future traffic condition for road user decisionTherefore the aim of this special issue is to introduce forthe readers a number of papers on various aspects of trafficinformation management

Topics covered in this issue include three main parts(1) traffic information estimation and prediction (2) trans-portation safety and security and (3) logistics transportation

traffic management This special issue has received a totalof 32 submitted papers with only 5 papers accepted A highrejection rate of 8438 of this issue from the review processis to ensure that high-quality papers with significant resultsare selected and published The three topics and acceptedpapers are briefly described below

2 Traffic Information Estimation andPrediction

Papers on analytical methods for traffic information estima-tion and prediction are as follows (1) ldquoA Method for DrivingRoute Predictions Based on HiddenMarkovModelrdquo by N Yeet al and (2) ldquoDetecting Traffic Anomalies in Urban AreasUsing Taxi GPS Datardquo by W Kuang et al

N Ye et al fromChina and SouthAfrica in ldquoAMethod forDriving Route Predictions Based on Hidden Markov Modelrdquoproposed a driving route predictionmethod based on hiddenMarkovmodel (HMM) to predict the traffic condition of eachroad segment for driverrsquos reference Furthermore amethodoftraining set extension based onK-means++ and a smoothingtechnique was used to build the HMM for route predictionsIn their experimental environment several training and testexamples in Jiangsu China were selected to evaluate theirproposed method The experimental results illustrated that

Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2015 Article ID 613940 2 pageshttpdxdoiorg1011552015613940

2 Mathematical Problems in Engineering

the correct prediction rate of their proposed method couldbe high

W Kuang et al from China in ldquoDetecting Traffic Anoma-lies in Urban Areas Using Taxi GPS Datardquo proposed atraffic anomalies detection method which could combine thewavelet transformmethod and principal component analysis(PCA) to detect traffic anomalies Moreover their proposedmethod could estimate and obtain information regardingthe spatial distribution of traffic flows In their experimentalenvironment several taxicabs collected and reported theirGPS data in Harbin China for the evaluation of theirproposed method The experimental results indicated thata number of the traffic anomalies could be detected andreported for managers to solve traffic jam

3 Transportation Safety and Security

Paper on analytical methods for transportation safety andsecurity is presented as follows S-K Ryu et al from Koreain ldquoImage-Based Pothole Detection System for ITS ServiceandRoadManagement Systemrdquo proposed a pothole detectionmethod based on various features in two-dimensional (2D)images which included three steps (1) segmentation based onHistogram Shape-Based Thresholding (HST) (2) candidateregion extraction in accordance with various features (egsize and compactness) and (3) decision by comparing pot-hole and background features In their experimental environ-ment several video clips in Korea were selected to evaluatetheir proposedmethodThe experimental results showed thatthe accuracy precision and recall of their proposed methodwere higher than previous methods

4 Logistics Transportation TrafficManagement

Papers on analyticalmethods for logistics transportation traf-fic management are as follows (1) ldquoIdentifying Key Factorsfor Introducing GPS-Based Fleet Management Systems tothe Logistics Industryrdquo by Y-C Hu et al and (2) ldquoNovelEncoding and Routing Balance Insertion Based ParticleSwarm Optimization with Application to Optimal CVRPDepot Location Determinationrdquo by R-M Chen and Y-MShen

Y-C Hu et al from Taiwan in ldquoIdentifying Key Factorsfor IntroducingGPS-Based FleetManagement Systems to theLogistics Industryrdquo combineddecision-making trial and eval-uation laboratory (DEMATEL) and analytic network process(ANP) to determine the key indicators (eg funding andbudget experience and ability of consultants project teamcomposition user recognition timely and correct informa-tion and degree of completeness of transmission equipment)for introducing GPS-based fleet management systems totransport companies In their experimental environmenta transport company in Taiwan was selected to evaluatetheir proposed method The experimental results indicatedthat adequate annual budget planning enhancement of userintention and collaboration with consultants were the keyindicators for successfully introducing the systems

R-M Chen and Y-M Shen from Taiwan in ldquoNovelEncoding and Routing Balance Insertion Based ParticleSwarm Optimization with Application to Optimal CVRPDepot Location Determinationrdquo proposed a hierarchicalparticle swarm optimization (PSO)with two layers (ie outerlayer PSO and inner layer PSO) for the establishment ofthe optimal depot location and the minimized total distanceof vehicle routing In their experimental environment nineinstances were selected from an accessible and credibledatabase which was designed by Augerat for the evaluationof vehicle routing algorithm The experimental results illus-trated that the costs of finding the new plant location andvehicle routing distance in a real world case could be reduced

Chi-Chun LoKuo-Ming ChaoHsu-Yang KungChi-Hua ChenMaiga Chang

References

[1] K Boriboonsomsin M J Barth W Zhu and A Vu ldquoEco-routing navigation system based on multisource historical andreal-time traffic informationrdquo IEEE Transactions on IntelligentTransportation Systems vol 13 no 4 pp 1694ndash1704 2012

[2] X Ma J Zhang X Yin and K S Trivedi ldquoDesign andanalysis of a robust broadcast scheme for VANET safety-relatedservicesrdquo IEEETransactions onVehicular Technology vol 61 no1 pp 46ndash61 2012

[3] A Bazzi B M Masini and O Andrisano ldquoOn the frequentacquisition of small data through RACH in UMTS for itsapplicationsrdquo IEEE Transactions on Vehicular Technology vol60 no 7 pp 2914ndash2926 2011

[4] K Zheng F Liu Q Zheng W Xiang and W Wang ldquoA graph-based cooperative scheduling scheme for vehicular networksrdquoIEEE Transactions on Vehicular Technology vol 62 no 4 pp1450ndash1458 2013

[5] B-F Wu and J-H Juang ldquoAdaptive vehicle detector approachfor complex environmentsrdquo IEEE Transactions on IntelligentTransportation Systems vol 13 no 2 pp 817ndash827 2012

[6] B Tian B T Morris M Tang et al ldquoHierarchical and net-worked vehicle surveillance in ITS a surveyrdquo IEEE IntelligentTransportation Systems Magazine vol 16 no 2 pp 557ndash5802015

[7] M-F Chang C-H Chen Y-B Lin and C-Y Chia ldquoThefrequency of CFVD speed report for highway trafficrdquo WirelessCommunications and Mobile Computing vol 15 no 5 pp 879ndash888 2015

Research ArticleNovel Encoding and Routing Balance Insertion Based ParticleSwarm Optimization with Application to Optimal CVRP DepotLocation Determination

Ruey-Maw Chen1 and Yin-Mou Shen2

1Department of Computer Science and Information Engineering National Chin-Yi University of Technology Taichung 41170 Taiwan2Department of Information Management Kun Shan University Tainan 710 Taiwan

Correspondence should be addressed to Ruey-Maw Chen raymondncutedutw

Received 21 November 2014 Revised 10 April 2015 Accepted 15 April 2015

Academic Editor Kuo-Ming Chao

Copyright copy 2015 R-M Chen and Y-M ShenThis is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited

A depot location has a significant effect on the transportation cost in vehicle routing problems This study proposes a hierarchicalparticle swarm optimization (PSO) including inner and outer layers to obtain the best location to establish a depot and thecorresponding optimal vehicle routes using the determined depot locationThe inner layer PSO is applied to obtain optimal vehicleroutes while the outer layer PSO is to acquire the depot location A novel particle encoding is suggested for the inner layer PSOthe novel PSO encoding facilitates solving the customer assignment and the visiting order determination simultaneously to greatlylower processing efforts and hence reduce the computation complexity Meanwhile a routing balance insertion (RBI) local searchis designed to improve the solution quality The RBI local search moves the nearest customer from the longest route to the shortestroute to reduce the travel distance Vehicle routing problems from an operation research library were tested and an average of 16total routing distance improvement between having and not having planned the optimal depot locations is obtained A real worldcase for finding the new plant location was also conducted and significantly reduced the cost by about 29

1 Introduction

The vehicle routing problem (VRP) is a scheduling problemencountered in logistic arrangement an extension of thetraveling salesman problem As different restrictions (vehiclecapacity limits visit time limits goods pick- and deliverydemands etc) there are also dissimilar types of VRPs suchas capacitated VRPs (CVRPs) involving only vehicle capacitylimits capacitated VRPs with time windows involving bothvehicle capacity and visit time limits at the same timeVRPs with pickups and deliveries involving pickup anddelivery demands multiple depot VRPs involving multipledepots and periodic VRPs involving customs with periodicdemands This study focuses on capacitated vehicle routingproblems In operation research vehicle routing problemshave been confirmed to be NP-hard Accurate optimal solu-tions to this type of problem can be obtained with exactalgorithms [1] within a limited time only when the problemscale is small With problems of a larger scale the amount

and time of calculation required make it impossible to obtainoptimal solutionswith exact algorithmswithin a limited timeFor this reasonmany researchers have come upwith a varietyof heuristic and metaheuristic methods in recent years tocope with vehicle routing problems including the evolutioncomputation memetic algorithm genetic algorithm (GA)local search metaheuristic artificial bee colony algorithmant colony optimization (ACO) and particle swarm opti-mization (PSO) Prins [2] used two memetic algorithmsfor heterogeneous fleet vehicle routing problems Repoussiset al [3] applied a hybrid evolution strategy for the openvehicle routing problem Gajpal and Abad [4] proposeda saving-based algorithm for vehicle routing problem inwhich a new route is created by merging two existing routesMunawar et al suggested a cellular genetic algorithm withlocal search to solve CVRP [5] Pop et al integrated a GAwith a local search to globalize the approach to the CVRP [6]In [7] a local search metaheuristic including the static movedescriptor strategy for exploration and the promises concept



for avoiding search cycling and inducing diversification wasdesigned for the VRP with simultaneous pick-ups and deliv-eries Fleszar et al proposed an effective variable neighbor-hood search scheme based on reversing the routing segmentand exchanging routing segments for solving the openVRP tominimize the number of vehicles as well as the total travelleddistance [8] Meanwhile an adaptive variable neighborhoodsearch together with diversification local search methodswas utilized to investigate the homogeneous fleet VRP [9]Artificial bee colony algorithm with a local optimizationstrategy based on a scanning strategy for an open VRP wasstudied by Yao et al [10] Szeto et al also applied an enhancedversion of artificial bee colony for solving the CVRP [11]Ant colony optimization is a well-known metaheuristic forcombinatorial optimization problems An ant colony systembased algorithm was proposed by Favaretto et al [12] tosolve VRP with multiple time window constraints Yu et alrecommended an improved ACO which implements a newant-weight strategy to update the increasing trail pheromoneand a mutation operation to solve VRP [13] A PSO-basedscheme with two solution encodings and the correspondingdecodings for solving CVRP was investigated by Ai andKachitvichyanukul [14] In [15] a PSO-based approach inwhich a variable neighborhood descent local search is per-formed to solve the VRPwith pickup and delivery at the sametime Meanwhile Marinakis et al [16] proposed a hybridalgorithm based on PSO for solving VRP with stochasticdemand Moreover a VRP with fuzzy demands was solvedby applying a PSO-based approach in which a novel encodingmethod was introduced [17]

Among them PSO has the advantage of requiring lessparameters and faster convergence rates and has thereforebeen adopted by many researchers to solve various problemsAbido [18] employed PSO to solve the optimal setting ofpower flow Kang andHe [19] proposed a novel discrete parti-cle swarm optimization algorithm for meta-task assignmentin heterogeneous computing systems and used a migrationmechanism to escape from possible local optimum A flowshop sequence dependent group scheduling problem wasresolved using PSO based on a ranked order value encodingscheme [20] Meanwhile Chen [21] presented PSO with jus-tification technique integrated to solve resource-constrainedproject scheduling problems Moreover an application ofPSO to solve task-resource assignment in a heterogeneousgrid was provided by Chen and Wang [22] AdditionallyChen and Sandnes [23] applied constriction PSO to solveman-day scheduling problems

Scholars have established different restriction databasesto help solve VRP problems but the objectives are mostlyto plan the least costly vehicle routes when the locations ofdepots and customers are already known A dynamic VRPwhich considers new customer requests while the vehiclerouting is in progress was also investigated by using PSO[24] In some industries 25 of the companyrsquos total revenuemust be used to pay for materials delivery as well as shippingcosts to ship products Restated the transportation cost isan extremely important consideration for many businessesTherefore efficient vehicle routing is crucial Meanwhile siteselection has a significant impact on the fixed and changing

costs and the impact of the companyrsquos risk and profits Hencesetting the operating site location is one of themost importantdecisions in many companies such as FedEx The goal of siteselection is to allow the company to reduce the transportationcost so as to get the most benefit Site selection can beany operating site selection including VRP depot locationselection However most studies focus on solving VRP basedon fixed depots In logistic businesses besides fine vehicleroute planning good choice of depot locations is also animportant issue to reduce business costs and hence increaseprofits Restated solving both the optimal depot location aswell as the optimal vehicle routes is necessary Thereforethis investigation focuses on solving these two issues by ahierarchical PSO involving two PSO algorithms one for theinner layer and the other for the outer layer The outer-layer PSO is first applied to establish the optimal depotlocation then the inner PSO is used to produce the optimalvehicle routing This optimal routing involves the customer-to-vehicle assignment and visit order determination issuesThese two issues are commonly resolved by two separatePSOs in most studies hence much effort is required There-fore a novel particle encoding scheme is proposed to dealwith those two issues simultaneously to greatly reduce theprocessing effort Meanwhile a new local search strategy isalso designed and employed to improve solution qualityThisnew designed local search is named routing balance insertion(RBI) local search herein it is inspired by the well-usednearest neighborhood heuristic in TSP The RBI local searchselects the nearest customer on the longest routing clusterand inserts the selected node into the shortest routing clusterto reduce the total travel distance The nearest customer isdetermined based on the distance between the customer andthe centroid of the shortest routing cluster

The organization of this work is as follows Section 2describes the interested capacitated vehicle routing problemsThe proposed scheme including novel particle encoding androuting balance insertion local search is given in Section 3Section 4 demonstrates the experimental results and analysisFinally conclusions are made in Section 5

2 Problem Description

The vehicle routing problem was first proposed by Dantzigand Ramser in 1959 [25] It was very similar to the conceptof distribution of goods by logistic businesses in reality Theproblem involved the demands of each of many customersscattered about different places The depot had to assignvehicles to visit (service) all the customers and satisfy theirneeds by planning the shortest total travel distance withoutviolating any restrictions

In a CVRP there are a fixed number of customers anda depot The locations of each customer and the depot areknown (indicated with Cartesian coordinates) Set C =

1198881 1198882 119888

119899 stands for the set customers 119888

1 1198882 119888

119899are

the customers The depot will send out a fleet comprisingseveral vehicles The vehicle fleet V = V

1 V2 V

119896 in

which 119896 is the number of vehicles Each customer has adifferent cargo demand and each vehicle has a carryingcapacity limitation Each vehicle must leave from the depot

Mathematical Problems in Engineering 3

Custo

mer

-veh

icle

assig

nmen

t

Opt

imiz

ed as

signm

ent

CV

c1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

CV

c1c2

cn

12

k

middot

Figure 1 Customer-to-vehicle assignment

and return to the depot at the end Each customer has to bevisited once and once only The objectives and restrictions ofthe CVRP are then defined as follows

Fitness = min119899

sum

119894=0

119899

sum

119895=0

119896

sum

V=1119889119894119895119883

V119894119895+ 1198891198990119883

V1198990

119894 = 119895 (1)

119899

sum

119894=0

119899

sum

119895=0

119883

V119894119895119903119894le 119876V 119894 = 119895 V isin 119881 (2)

119883

V119894119895

=

1 a customer 119894 to 119895 is on the route of vehicle V

0 otherwise

(3)

In (1) the objective function of the VRP is defined asto obtain the shortest total travel distance The 119889

119894119895is the

distance from the customer 119894 to customer 119895 and 119883V119894119895stands

for whether vehicle V will go from customer 119894 to customer 119895When 119883V

119894119895= 1 it means vehicle V travels from a customer

119894 to 119895 On the other hand when 119883V119894119895= 0 vehicle V does

not travel from customer 119894 to customer 119895 In (2) the totaldemands from customers served by vehicle Vmay not exceedthe carrying capacity of vehicle V The 119903

119894stands for the cargo

demand of customer 119894 while 119876V is the maximum carryingcapacity defined for vehicle V The objective is to obtain theshortest total travel distance but each vehicle may not violatethe maximum capacity restriction throughout the tour

This investigation is interested in determining the optimaldepot location as well as the optimal vehicle routing Thisproblem to obtain the optimal vehicle routes first needsallocation of the 119899 customers to 119896 vehicles Hence there isa surjection from customer collection C = 119888

1 1198882 119888

119899 to

vehicle collection V = V1 V2 V

119896 that is customer to

vehicle assignment as shown in Figure 1 Next determinationof the optimal visit order for each vehicle is needed asdisplayed in Figure 2

To acquire optimal customer-to-vehicle assignment andoptimal visit order for each vehicle a particle swarm opti-mization (PSO) with a novel particle encoding scheme is pro-posed to resolve these two issues at the same time Restated

with the help of the novel particle encoding scheme thecustomer assignment and the visiting order determinationcan be solved concurrently

Meanwhile a depot has a very significant effect on thetransportation cost Therefore a hierarchical PSO is utilizedthe position of the depot is adjusted with the outer PSOand then the inner PSO is applied to determine the optimalcustomer assignment and optimal visit order with minimumtotal vehicle routes

3 Particle Swarm Optimization withProposed Designs

This study focuses on applying hierarchical PSO to obtainoptimal depot location as well as the optimal vehicle routesIn this Section PSO is first introduced next a novel particleencoding for the inner and outer layer PSOs are presentedTo enhance the PSO performance routing balance insertionlocal search is designed

31 Particle SwarmOptimization (PSO) Particle swarm opti-mization is a type of collective intelligence It was first putforward in 1995 by Kennedy and Eberhart [26] who wereinspired by the group behavior of biological creatures lookingfor food together In the operation of a PSO algorithm theposition of a particle stands for the solution to the problemIn PSO a particle moves in the solution space and usestwo experiences as references for further motion namelythe optimal individual experience and the optimal groupexperience The optimal group experience indicates that theentire group has been placed in the best position and theoptimal individual experience means each particle has beenplaced in its best position When calculating the newmovingspeed of a particle in each iteration besides the original speedthe positions of the optimal group experience and the optimalindividual experience are also referred to Suppose that an119873 number of particles are scattered in an 119871-dimensionalspace The position vector of the 119894th particle (119894 = 1 119873)is composed of 119871 vector components 119883

119894= 119883

1198941 119883

119894119871

indicates the position vector of particle 119894 in which119883119894119895stands

for the 119895th vector component of the 119894th particle The velocityvector of the 119894th particle is also composed of 119871 components119881119894= 1198811198941 119881

119894119871 The optimal individual experience of the

119894th particle is thus represented as 119875119894= 1198751198941 119875

119894119871 whereas

the optimal swarm experience (119866best) is 119866 = 1198661 119866

119871

These velocity and position update rules are shown below

119881

new119894119895

= 119908 times 119881119894119895+ 1198881times 1199031times (119875119894119895minus 119883119894119895) + 1198882times 1199032

times (119866119895minus 119883119894119895)

119883

new119894119895= 119883119894119895+ 119881

new119894119895

(4)

In (4) 119908 is the inertia weight used to determine thelevel of effect of the previous velocity on the new velocityIn PSO algorithms inertia weight is an important factorthat has influence on the search ranges of particles When119908 increases the searching movement of a particle is broaderand global exploration is suitable On the other hand when


1

Depot

310

8

2

95

7

6

4

Opt

imiz

ed sc

hedu

le

Opt

imiz

ed as

signm

ent

1

Depot

72

8

10

95

3

6

4

7

Depot

310

8

5

92

1

6

4

CV

c1c2

cn

12

k

middot

Figure 2 Visit order optimization

Table 1 Novel compound particle encoding (inner layer PSO)

Index 1 2 sdot sdot sdot 119899 119899 + 1 119899 + 2 sdot sdot sdot 119899 + 119896 minus 1

119883

119881

119894119883

119881

1198941119883

119881

1198942sdot sdot sdot 119883

119881

119894119899119883

119881

119894119899+1119883

119881

119894119899+2sdot sdot sdot 119883

119881

119894119899+119896minus1

Key Cus1 Cus2 sdot sdot sdot Cus119899

Veh1 Veh2 sdot sdot sdot Veh119896minus1

the search space is narrower local exploitation will be moreappropriate Therefore proper adjustment of 119908 to balanceglobal exploration and local exploitation is required andimportant Meanwhile 119888

1and 1198882are learning factors which

have an effect on particlesrsquo learning of global experience andindividual experience whereas 119903

1and 1199032represent random

numbers within [0 1]

32 Novel Particle Encoding for Inner Layer PSO The par-ticle position vector represents the solution of a studiedproblem and the particle position encoding is the corestep in PSO Before the inner layer PSO performs visitorder decision-making and fitness calculations the positionvector (119883119881

119894) has to be converted into the visit sequence of

a vehicle Restated each customer the vehicle is assignedto have to be determined before an assessment can beconducted Hence to facilitate finding the optimal solutionand reduce the processing effort this work designs a novelcompound particle encoding scheme to reduce the customer-to-vehicle assignment and visit order determination effortfor the inner layer PSO Herein a particle of the inner-layerPSO includes customers and vehicles assigned as shown inTable 1 In Table 1 the position vector includes 119899 + (119896 minus1) components that is 119883119881

119894= 119883

119881

1198941 119883

119881

119894119899 119883

119881

119894119899+119896minus1

Meanwhile each component is associated with a key(Key = Cus

1Cus2 Cus

119899Veh1Veh2 Veh

119896minus1) For

customer-to-vehicle assignment 119899 customers are to beassigned to 119896 vehicles that is 119899 customers can be regardedas being clustered into 119896 groups Therefore (119896 minus 1) dividingpoints are needed that is the reason Veh

1ndashVeh119896minus1

(119896 minus 1components) are added

The visit sequence of each vehicle and each customer avehicle is assigned to are determined simultaneously by using

a random key scheme Take six customers and three vehiclesfor example Figure 3 shows a solution (119883119881

119894) obtained with

PSO The components of the position vector are sorted inascending order then the key values are rearranged accord-ing to the sorted values of119883119881

119894to generate a key sequence set

This key sequence is then defined as the vehicle assignmentwith the Veh

119895as the dividing point Restated all customers

before the dividing point Veh1are assigned to vehicle 1 all

customers between Veh1and Veh

2are assigned to vehicle 2

and so forth Finally customers after Veh119896minus1

are assigned tovehicle 119896Moreover the customers visit sequence for a vehicleis then defined as the visiting order for that vehicle Thetotal travel distance can then be calculated according to (1)after the vehicle assignment and visiting order are resolvedFor example customers 1 2 and 5 are assigned to vehicle 2and the visiting order for vehicle 2 would be from customer2 to customer 5 then customer 1 as indicated in Figure 3Hence the proposed novel PSO encoding scheme in innerlayer PSO can facilitate solving the customer assignment andthe visiting order determination at the same time to greatlylower processing effort and hence reduce the computationalcomplexity

33 Particle Encoding for the Outer Layer PSO The particleencoding for the outer layer PSO solutions is conductedby using a position vector consisting of two componentsrepresenting the 119883 and 119884 coordinates of the depot locationThe outer layer PSO solution (X119863 = 119883

119863

1 119883

119863

2) is shown

in Table 2 The fitness calculation is then performed bytransferring the depot coordinates (X119863) to the inner layerPSO for optimal routing calculation and the resulting totalrouting distance is adopted as the fitness value of the outerlayer PSO


Key2 13 08 24 19 02 12 21

02 08 12 13 19 2 21 24Key

Sorting in ascent order

Vehicle assignment

Visit order

Veh 1

Veh1

Veh1 Veh2

Veh2

Cus1

Cus1

Cus1

Veh 2

Cus2

Cus2

Cus2

Veh 3

Cus3

Cus3

Cus3

Cus4

Cus4

Cus4

Cus5

Cus5

Cus5

Cus6

Cus6

Cus6

XiV

XiV

Figure 3 The solution decoding process (inner layer PSO)

Table 2 Solution representation (outer layer PSO)

X119863 119883

119863

1119883

119863

2

Depot location 119883 coordinate 119884 coordinate

34 Routing Balance Insertion Local Search The local searchis a search tactic to generate new solutions in the neighbor-hood of the current solution to attempt to find a solution withbetter quality A new local search is designed and conductedto generate a new solution and is selected to be the startingpoint of the algorithm when the next iteration takes place ifit is a better solution

The new local search tactic named routing balance inser-tion (RBI) local search is applied in the inner layer PSOwhich is inspired from the well-used nearest neighborhoodheuristic in TSP The RBI local search moves the nearestcustomer from the longest route to the shortest route toreduce the travel distance the nearest customer is determinedbased on the distance between the customer and the centroidof the shortest routing clusterThe operations of the designedRBI local search are as follows

Step 1 Select the longest routing path and the shortestrouting path Figure 4 shows the resulting CVRP resultsRoute-1 is the routing path starting from depot (119874) andvisiting 119860 119861 119862 119863 119864 and 119865 then back to 119874 Route-2 isthe routing path starting from 119874 and visiting 119866 119867 and 119868then back to the depot Assuming the travel distances of thecorresponding vehicle routes are 1198891 1198892 and 1198893 respectivelySuppose the max1198891 1198892 1198893 is 1198891 and the min1198891 1198892 1198893 is1198892

Step 2 Calculate the centroid position of the customersconsisting of the shortest route (Route-2) The centroidposition (119862119862 = (119909

119862 119910119862)) can be yielded by

119909119862=

sum

119896

119894=1119909

V119894+ 119909119874

119896 + 1

119910119862=

sum

119896

119894=1119910

V119894+ 119910119874

119896 + 1

(5)

F

O

DE

G

HA

I

C

J

B

K

Route-1

Route-2

Route-3

Figure 4 Obtained CVRP results

F

O

DE

G

HA

I

C

J

B

K

dE

dF

dD

dC

dB

dA

CC

Figure 5 The centroid and the distances from customer on thelongest route

In (5) 119909119862and 119910

119862are the coordinates of the centroid position

of route V (vehicle V) The 119909V119894and 119910V

119894are the coordinates of

the customers assigned to the vehicle V 119909119874and 119910

119874are the

coordinates of the depot position

Step 3 Calculate the distances from the customers assignedto the longest route (Route-1) to the centroid Assuming119889119860 119889119861 and 119889119865 are the distances from customers 119860 119861 and 119865 to the centroid as displayed in Figure 5 Suppose 119889119861 isthe minimum distance that is customer 119861 is the nearest oneto the shortest route


F

O

DE

B

C

JK

G

H

I

A

(a) 1198891 = 119874119861 + 119861119866minus 119874119866

F

O

DE

B

C

JK

G

H

I

A

(b) 1198892 = 119866119861 + 119861119867minus 119866119867

F

O

DE

C

J

A

K

G

H

IB

(c) 1198893 = 119867119861 + 119861119868 minus 119867119868

F

O

DE

B

C

J

A

K

G

H

I

(d) 1198894 = 119868119861 + 119861119874minus 119868119874

Figure 6 Four possible insertion positions

Step 4 Delete customer 119861 from Route-1 and insert 119861 intoRouter-2The travel distance of theRoute-1 decreases after thecustomer 119861 is removed the decreased distance is 119889 = 119860119861 +119861119862 minus 119860119862 Meanwhile there are four possible positions forinserting 119861 as illustrated in Figure 6 The increased distancesafter inserting 119861 to the four possible positions are 1198891 =

119874119861 + 119861119866 minus 119874119866 1198892 = 119866119861 + 119861119867 minus 119866119867 1198893 = 119867119861 + 119861119868 minus119867119868 and 1198894 = 119868119861 + 119861119874 minus 119868119874 respectively The insertionposition is then determined by comparing 1198891 1198892 1198893 and1198894 Restated the insertion position decision is based on themin1198891 1198892 1198893 1198894 For example the customer 119861 is beinginserted between119866 and119867 if the 1198892 is theminimum increaseddistance as in Figure 6(b)

35 Optimal Depot Location Determination The optimaldepot location is determined using the outer layer PSO Thedetermined particle solution X119863 is passed to the inner layerPSO as the depot location The inner layer PSO solves theCVRP problem on the basis of this depot location and theminimum total vehicle routing distances (Fitness in (1)) arereturned to the outer PSO This returned Fitness is thenused as the objective corresponding to X119863 Accordinglyparticle experience and swarm experience can be obtainedThereafter the velocity in the outer layer PSO is updateda new position X119863 is generated and will be the new depotlocation After alternating evolutions of the inner layer andouter layer PSO an optimal depot location can be acquired

36 Hierarchical PSO The collaboration operation of theproposed inner and outer layer PSOs is as follows

(1) Outer layer PSO outputs determined depot location(X119863) to the inner layer PSO

(2) Inner layer PSO determines total travel distance(TTD) based on X119863 returns the total travel distanceto the outer layer PSO

(3) Outer layer PSO

(i) evaluates the quality of the depot location (X119863)that is fitness(X119863) = TTD

(ii) updates individual and swarm experience(iii) updates velocity and position vector(iv) outputs new depot location (X119863) to the inner

layer PSO

(4) Repeats Steps 3 and 4 until termination condition ismet

(5) Outer layer PSO outputs the optimal depot locationand the corresponding vehicle routes

The detailed flowchart of the proposed hierarchical PSO foroptimal CVRP depot location and optimal vehicle routes issummarized in Figure 7


Start

Termination condition met


Output optimal depot location and optimal vehicle routing

End

Yes Yes

NoNo

YesNo

Inner layer Outer layer

Initialize VVX

V

Update VVX

V

Initialize VDX

D

Update VDX

D

search(XV)

Fitness(X ) lt

Fitness(XV)

Update(SA)

Fitness( )

Updateand

Pass XD

to inner layer PSO

Fitness(XD) =

Fitness( )= XLSV

GVbest

XVnew

PVbest

XVnew X

Vnew

Updateand

GVbest

PVbest

GVbest

LSV

XVLS = local

Figure 7 Flowchart of the proposed hierarchical PSO

Table 3 Complexity of the VRP scheduling problem

Customers Vehicles Solution space119899 = 119883119883 minus 1 119898 119898 times (119899119898) times 119898

119899

31 5 5 times 6 times 531 asymp 167 times 1025



4 Experimental Results

To verify the performance of the method proposed in thiswork to establish the optimal depot location simulations ona famous benchmark were conducted The instances testedare those designed by Augerat aiming at capacitated vehiclerouting problems There are 9 instances selected from thedatabase at httpwwwbranchandcutorgVRPdata they areA-n32-k5 A-n33-k5 A-n36-k5 A-n45-k6 A-n45-k7 A-n55-k9 A-n60-k9 A-n62-k8 and A-n64-k9 An instance isexpressed by A-n119883119883-k119884 where119883119883 stands for the number ofcustomers plus depots and119884 indicates the number of vehicles

Table 3 demonstrates the difficulty of solving the studiedCVRP problems Assuming 119899 customers are serviced by119898 vehicles in average every vehicle needs to visit 119899119898customers Therefore the time required by exhaustive search

Table 4 Particle complexity on finding optimal routes

Two PSOs Proposed PSONumber of component 119899 + 119899 119899 + (119898 minus 1)ExampleA-n32-k5 31 + 31 31 + 4

A-n54-k9 53 + 53 53 + 8

A-n64-k8 63 + 63 63 + 7

for the A-n32-k5 instance would be 167 times 1025 times 10minus8seconds asymp 19 times 1012 days with a solution that can be found in001 120583sec (10minus8 sec) is assumed For another example case thetime required by exhaustive search for the A-n64-k8 instancewould be 253times 1062 times 10minus8 secondsasymp 369times 1049 days Hencea PSO metaheuristic algorithm is applied in this study

Table 4 lists the required number of component velocityand position vectors for the inner PSO to find the optimalroutes To solve the two issues encountered in obtainingthe CVRP optimal routes there is one commonly useddesign when applying PSO two PSOs are dedicated tosolve corresponding issues However the required numberof components in either the velocity or position vector is119899 + 119899 components in total however only 119899 + (119898 minus 1)

components are required in the proposed novel particle


encoding scheme Hence the computational complexity isdecreased dramatically for large scale problems

In this work the experiments were processed in twostages The first stage is to find out the best mechanismsemployed in the inner layer PSO including the local searchThe second stage is to check the improvements when thedepot location is determined by using the outer layerPSO Restated the resulting fitnesses after and before outerlayer PSO application are compared to observe the level ofimprovement During the test in the first stage the customersprovided in the benchmark were divided into small mediumand large scales Three instances for each scale were adoptedto run the test The inner layer PSO parameters were 100particles the learning factors 119888

1= 2 and 119888

2= 1 and the

number of iterations was 1000 The outer layer PSO involved8 particles the learning factors were set to 119888

1= 1198882= 2 and 100

iterations were conductedThe comparison criterion is on thebasis of deviation The deviation (DEV) is defined in

DEV () =Makespansol minus BKS

BKStimes 100 (6)

where BKS is the best known solution provided in thebenchmarkMakespansol is the shortest total routing distanceobtained by the proposed method The best deviation from10 trials was selected for comparison Moreover the averagedeviation (Avg Dev) is also defined as in

Avg Dev () =sum

119899

119894=1DEV119894

119899

(7)

where 119899 is the trial runs for a specific test problem instance10 trial runs were conducted in this work that is 119899 = 10

The testing environment of the experiment included theWindows 7 SP1 operating system running on an Intel Core i7CPU 4770 340GHz CPU with 4GB RAM C was applied toimplement the method proposed in this study

41 Inner-Layer PSO Local Searches To test the efficiencyof different local searches interchange (LS

1) RBI (LS

2)

combining interchange and RBI (LS3) were tested The

results are as shown in Figure 8 It indicates that either swapor RBI local search is able to improve the efficiency Theproposed RBI local search (Avg Dev = 18) outperformsswap local search (Avg Dev = 20) and without the localsearch (Avg Dev = 28) Moreover both swap and RBIinvolved in the algorithm are able to further enhance theperformance (Avg Dev = 14) Therefore the inner layerPSO involving swap local search and RBI local search wasincluded while searching for the optimal depot location bythe outer layer PSO

42 Outer Layer PSO In this section the experimentalresults with and without applying the outer layer PSOto find the optimal depot location are compared Thedepot locations provided in the benchmark were used asthe default depot locations the fitness (Fit) based on (1)was calculated Figure 9 shows the inner layer PSO andouter layer PSO evolution curves for the A-32-k5 instance

0102030405060708090

Aver

age d

evia

tion

()

A-n3

2-k5

A-n3

3-k5

A-n3

6-k5

A-n4

5-k6

A-n4

5-k7

A-n5

5-k9

A-n6

0-k9

A-n6

2-k8

A-n6

4-k9

Aver

age

wo LSLS1

LS2LS3

Figure 8 Simulation results of applying local searches

Figures 10(a) and 10(b) display the resulting vehicle routesbefore and after applying outer layer PSO respectively Thefitness of using the default depot is 784 but the fitness ofusing a determined depot by the proposed outer layer PSOis 660 Restated the determined depot would greatly reducethe vehicle routing cost

Table 5 displays the experimental results of using defaultdepot location (without adjustment of the depot locationie before the outer layer PSO was applied) and determineddepot location (with adjustment of the depot location afterouter layer PSO application) Ten trials were conducted theminimum fitness (Min Fit) and average fitness (Avg Fit)are provided Meanwhile the improvement was calculatedaccording to

Imp() =Fitness

119908119900minus Fitnessdepot

Fitness119908119900

times 100 (8)

where Fitness119908119900

is the fitness without the depot locationadjustment and the Fitnessdepot is the fitness with thedepot location adjustment Restated the Imp represents thepercentage of the reduced fitness (total routing distancedecreased) According to the experimental results up to18 average minimum Imp (Min Imp) and 16 averagedImp (Avg Imp) of trial runs were acquired Therefore theproposed scheme in this work is able to additionally allowcompanies to determine the optimal depot or plant sitesetting

Finally a real world case was implementedThe real worldcase includes 15 cooperation factories and a new assemblyplant is planned to set up to produce commodities Thelocation of this assembly plant needs to be determined toreduce the costs The requirement is that the assembly plantneeds to send out 3 trucks to carry all needed parts fromall cooperation factories and back to the assembly plant forfurther processes The vehicle routing based on the originalplant location is displayed in Figure 11(a) the vehicle routingon the basis of the determined new plant location usingthe proposed scheme is illustrated in Figure 11(b) The travel


Fitn

ess

950

900

850

800

750

700

Iterations

0

50

100

150

200

250

300

350

400

450

500

550

600

650

700

750

800

850

900

950

1000

(a)

Fitn

ess

830

810

790

770

750

730

710

690

670

650

Iterations

0 5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

(b)

Figure 9 PSO evolution example for instance A-32-k5 (a) inner layer PSO and (b) outer layer PSO

(a) (b)

Figure 10 Resulting vehicle routes example for case A-32-k5 (a) without depot determination and (b) with depot determination by outerlayer PSO

Table 5 Improvement of the proposed scheme

Instance Default Determined depot ImprovementMin Fit Min Fit Avg Fit Min Imp Avg Imp

A-n32-k5 784 660 660 19 19A-n33-k5 661 627 632 5 5A-n36-k5 799 685 696 17 15A-n45-k6 944 842 931 4 1A-n45-k7 1146 829 864 38 33A-n55-k9 1073 1063 1078 1 0A-n60-k9 1408 1096 1118 28 26A-n62-k8 1315 1187 1098 19 18A-n64-k9 1177 1140 1081 33 30Average 18 16


(a) (b)

Figure 11 Vehicle routes based on (a) original plant location and (b) determined new plant location by the proposed PSO scheme

distances of the original plant vehicle routes and new plantvehicle routes are about 522 Km and 371 Km respectively

5 Conclusions

This study proposes a hierarchical PSO consisting of an innerlayer PSO and an outer layer PSO to obtain the optimal depotlocation and the corresponding vehicle routing to minimizethe total routing distance The inner layer PSO is used tofind the optimal vehicle routing while the outer layer is usedto determine the optimal depot location In the inner layerPSO a new designed routing balance insertion (RBI) localsearch is suggested to improve solution quality The RBIlocal search moves the nearest customer from the longestroute to the shortest route to reduce the travel distance thenearest customer selection is based on the distance betweena customer and the centroid of the shortest routing clusterThe experimental results with and without local searchschemes are demonstrated in Figure 8 in which the averagedeviation can be lowered (Avg Dev = 14) while applyinglocal searches Meanwhile a novel particle encoding schemeis designed to handle customer-to-vehicle assignment andcustomer visiting order issues simultaneously to greatlylower processing efforts and hence reduce the computationalcomplexity as indicated in Table 4

The experimental results indicate that the total vehi-cle routing distance of the tested instances is significantlyreduced up to an average improvement of 16 In the A-n45-k7 instance the minimum and average fitnesses of ten trialscan be improved up to 38 and 33 respectively Thereforethe location of a depot can indeed affect vehicle routing costswhich can be greatly lowered by the proposed hierarchicalPSOwith the novel encoding scheme and the RBI local searchin this study Restated the suggested PSO is able to effectivelyestablish the optimal location to set up a depot thus increas-ing profits According to the real-world case simulation asindicated in Figure 11 the new plant location is able to signif-icantly reduce the cost ((522 minus 371)522) times 100 cong 29

However to further enhance the performance local searchheuristics such as insertion exchange and other localsearches can be integrated into the proposed scheme Mean-while different metaheuristic algorithms such as geneticalgorithmand ant colony optimization can be utilized to solvethis studied problem in the future

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgment

This work was partly supported by the National ScienceCouncil Taiwan under ContractMOST 103-2221-E-167-009

References

[1] R Fukasawa H Longo J Lysgaard et al ldquoRobust branch-and-cut-and-price for the capacitated vehicle routing problemrdquoMathematical Programming vol 106 no 3 pp 491ndash511 2006

[2] C Prins ldquoTwo memetic algorithms for heterogeneous fleetvehicle routing problemsrdquo Engineering Applications of ArtificialIntelligence vol 22 no 6 pp 916ndash928 2009

[3] P P Repoussis C D Tarantilis O Braysy and G Ioannou ldquoAhybrid evolution strategy for the open vehicle routing problemrdquoComputers amp Operations Research vol 37 no 3 pp 443ndash4552010

[4] Y Gajpal and P Abad ldquoSaving-based algorithms for vehiclerouting problem with simultaneous pickup and deliveryrdquo Jour-nal of the Operational Research Society vol 61 no 10 pp 1498ndash1509 2010

[5] A Munawar MWahib M Munetomo and K Akama ldquoImple-mentation and Optimization of cGA+ LS to solve CapacitatedVRP over CellBErdquo International Journal of Advancements inComputing Technology vol 1 no 2 pp 16ndash28 2009


[6] P C Pop O Matei and C P Sitar ldquoAn improved hybridalgorithm for solving the generalized vehicle routing problemrdquoNeurocomputing vol 109 no 3 pp 76ndash83 2013

[7] E E Zachariadis and C T Kiranoudis ldquoA local searchmetaheuristic algorithm for the vehicle routing problem withsimultaneous pick-ups and deliveriesrdquo Expert Systems withApplications vol 38 no 3 pp 2717ndash2726 2011

[8] K Fleszar I H Osman and K S Hindi ldquoA variable neighbour-hood search algorithm for the open vehicle routing problemrdquoEuropean Journal of Operational Research vol 195 no 3 pp803ndash809 2009

[9] A Imran S Salhi andN AWassan ldquoA variable neighborhood-based heuristic for the heterogeneous fleet vehicle routingproblemrdquoEuropean Journal of Operational Research vol 197 no2 pp 509ndash518 2009

[10] B Yao P Hu M Zhang and S Wang ldquoArtificial bee colonyalgorithm with scanning strategy for the periodic vehiclerouting problemrdquo Simulation vol 89 no 6 pp 762ndash770 2013

[11] W Y Szeto Y Wu and S C Ho ldquoAn artificial bee colony algo-rithm for the capacitated vehicle routing problemrdquo EuropeanJournal of Operational Research vol 215 no 1 pp 126ndash135 2011

[12] D Favaretto E Moretti and P Pellegrini ldquoAnt colony systemfor a VRP with multiple time windows and multiple visitsrdquoJournal of Interdisciplinary Mathematics vol 10 no 2 pp 263ndash284 2007

[13] B Yu Z-Z Yang and B Yao ldquoAn improved ant colonyoptimization for vehicle routing problemrdquo European Journal ofOperational Research vol 196 no 1 pp 171ndash176 2009

[14] T J Ai and V Kachitvichyanukul ldquoParticle swarm optimizationand two solution representations for solving the capacitatedvehicle routing problemrdquo Computers amp Industrial Engineeringvol 56 no 1 pp 380ndash387 2009

[15] F P Goksal I Karaoglan and F Altiparmak ldquoA hybrid discreteparticle swarm optimization for vehicle routing problem withsimultaneous pickup and deliveryrdquo Computers amp IndustrialEngineering vol 65 no 1 pp 39ndash53 2013

[16] Y Marinakis G-R Iordanidou and M Marinaki ldquoParticleswarm optimization for the vehicle routing problem withstochastic demandsrdquoApplied SoftComputing Journal vol 13 no4 pp 1693ndash1704 2013

[17] Y Peng and Y-M Qian ldquoA particle swarm optimizationto vehicle routing problem with fuzzy demandsrdquo Journal ofConvergence Information Technology vol 5 no 6 pp 112ndash1192010

[18] M A Abido ldquoOptimal power flow using particle swarmoptimizationrdquo International Journal of Electrical PowerampEnergySystems vol 24 no 7 pp 563ndash571 2002

[19] Q Kang and H He ldquoA novel discrete particle swarm opti-mization algorithm for meta-task assignment in heterogeneouscomputing systemsrdquoMicroprocessors and Microsystems vol 35no 1 pp 10ndash17 2011

[20] D Hajinejad N Salmasi and R Mokhtari ldquoA fast hybridparticle swarm optimization algorithm for flow shop sequencedependent group scheduling problemrdquo Scientia Iranica vol 18no 3 pp 759ndash764 2011

[21] R-M Chen ldquoParticle swarm optimization with justificationand designed mechanisms for resource-constrained projectscheduling problemrdquo Expert Systems with Applications vol 38no 6 pp 7102ndash7111 2011

[22] R-M Chen and C-M Wang ldquoProject scheduling heuristics-based standard PSO for task-resource assignment in heteroge-neous gridrdquo Abstract and Applied Analysis vol 2011 Article ID589862 20 pages 2011

[23] R-M Chen and F E Sandnes ldquoAn efficient particle swarmoptimizer with application to man-day project schedulingproblemsrdquo Mathematical Problems in Engineering vol 2014Article ID 519414 9 pages 2014

[24] M R Khouadjia B Sarasola E Alba L Jourdan and E-GTalbi ldquoA comparative study between dynamic adapted PSO andVNS for the vehicle routing problem with dynamic requestsrdquoApplied Soft Computing vol 12 no 4 pp 1426ndash1439 2012

[25] G B Dantzig and J H Ramser ldquoThe truck dispatching prob-lemrdquoManagement Science vol 6 no 1 pp 80ndash91 19591960

[26] J Kennedy and R C Eberhart ldquoParticle swarm optimizationrdquoin Proceedings of the IEEE International Conference on NeuralNetworks vol 4 pp 1942ndash1948 Perth Australia December1995

Research ArticleA Method for Driving Route Predictions Based on HiddenMarkov Model

Ning Ye1 Zhong-qin Wang1 Reza Malekian2 Qiaomin Lin1 and Ru-chuan Wang1

1 Institute of Computer Science Nanjing University of Post and Telecommunications Nanjing 210003 China2Department of Electrical Electronic and Computer Engineering University of Pretoria Pretoria 0002 South Africa

Correspondence should be addressed to Reza Malekian rezamalekianupacza

Received 18 November 2014 Revised 4 January 2015 Accepted 21 January 2015

Academic Editor Chi-Hua Chen

Copyright copy 2015 Ning Ye et al This is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

We present a driving route prediction method that is based on HiddenMarkovModel (HMM)This method can accurately predicta vehiclersquos entire route as early in a triprsquos lifetime as possible without inputting origins and destinations beforehand Firstly wepropose the route recommendation system architecture where route predictions play important role in the system Secondlywe define a road network model normalize each of driving routes in the rectangular coordinate system and build the HMM tomake preparation for route predictions using a method of training set extension based on K-means++ and the add-one (Laplace)smoothing technique Thirdly we present the route prediction algorithm Finally the experimental results of the effectiveness ofthe route predictions that is based on HMM are shown

1 Introduction

Currently many drivers use different kinds of navigationsoftware to acquire better driving routes The main functionof vehicle route recommendation in the software is to findseveral routes between given origins and destinations bycombing some path algorithms with historical traffic datafor example Google Map and Baidu Map And then a drivercould select one of those recommendation routes accordingto personal preference driving distance and current roadcongestion information People usually would like to chooseroutes withmore smooth roads However the abovemethodsfor driving route recommendation have some problemsFirstly more people would like to choose routes with manysmooth road segments Thus the original relatively smoothroadswill become congested and the original congested roadswill become smooth Secondly once a route is selected thesoftware could not timely inform the driver to adjust theroute according to real-time traffic congestion data as the tripprogresses Finally most of traffic route navigation softwareprograms rely on historical data to predict traffic congestion[1] While some emergency situations arise for examplewhen organizing a large rally in an area a large number ofvehicles will move to this region in a short time leading to

traffic congestion in the area Obviously this case may nothave happened in previous historical data

In view of the above problems a driving route recom-mendation system is proposed and highlights a method fordriving route predictions based on the knowledge of HiddenMarkov Model (HMM) The method can predict which roadsegments are congested or smooth through route predictionsThe system will also update traffic information in real time inthe near future and inform the driver to adjust the drivingroute as the trip progresses

At present several methods of route prediction have beensuggested but there remain some problems Karbassi andBarth [2] described amethod to predict smart vehiclesrsquo routesbetween given starting and ending drop-off stations basedon a car-sharing application In our work the destinationnever needs to be inputted into the system beforehand Ourapproach also differentiates from the short-term route pre-diction in Krummrsquos work [3] Our method makes long-termpredictions about the entire route Froehlich and Krumm[4] found that a large portion of a typical driverrsquos trips arerepeated from the collected GPS data So based on this factthey predicted a driverrsquos entire route by using driversrsquo triphistory Simmons et al [5] firstly assumed that drivers havecertain routine routes and that by learning a model based on



previous experience one can accurately predict what a driverwill do in the future So based on this underlying premisethey presented an approach to predict driver intent usingHidden Markov Models However in fact it is impracticalto build a Hidden Markov Model for every driver and manyroutes are not fully regular When a driver takes a new routethe model for this driver could not predict the driverrsquos routeand destination intent

This paper is organized as follows The next sectiondescribes the architecture of our route recommendation sys-tem and explains each module in the system Section 3introduces how to construct a road network model andSection 4 presents how to define each of the driving routesbased on Section 3 The process of building HMM and themethod of making route predictions are discussed in Section5Then Section 6 shows experimental results Finally Section7 will conclude the paper

2 The Architecture of Driving RouteRecommendation System Based on HMM

The architecture of the driving route recommendation con-sists of the following phases (see Figure 1)

(i) Driving Route Predictions Based on HMM It is the core ofour recommendation system and is chiefly introduced in thispaper The module could find which routes a driver will beon when making a route prediction Even though we couldnot accurately gain the completely correct routes in practicethese possible routes are still very important for preestimatingtraffic congestion in the future

(ii) Traffic Congestion Preestimation It is mainly used topredict the congestion of each road At the time 119879119896 thecongestion level 119877119878(119879119896 119877119894) of each road 119877119894 is denoted by thetotal number of possible driving routes with the road 119877119894 ina time period The higher the value 119877119878(119879119896 119877119894) is the morecongested the road 119877119894 is

(iii) Vehicle Route Recommendation It collects informationabout just-driven road segments and traffic congestion sit-uations to introduce better routes for drivers based onexisting path algorithms [6ndash10] (all of these route planningalgorithms take traffic congestion situations into account inthe process of a vehicle route guidance) without presettingthe destination beforehand

(iv) HMMCorrection It is used to correct the HMMdepend-ing on new input driving routesThe given corpus of trainingsamples may not fully include all of possible driving routesWith the increase of inputting driving routes the amount oftraining data for training HMM will also grow which couldimprove the prediction accuracy

3 The Definition of Road Network Model

This section will give details on how to build a road networkmodel in the rectangular coordinate system The connectionrelationship between roads is followed strictly in the model

And it should reflect the difference between roads as large aspossible

Assume that each road 119877119894 is described as a line segment119877119894119909 perpendicular to 119909-axis that is the coordinate of twoendpoints of a line segment 119877119894119909 is separately defined by(1198831198941 1198841198941) and (1198831198941 1198841198942) where 1198841198941 = 1198841198942 or a line segment119877119894119910 perpendicular to 119910-axis that is the coordinate of twoendpoints of a line segment 119877119894119910 is separately defined by(1198831198941 1198841198941) and (1198831198942 1198841198941) where1198831198941 = 1198831198942

In the rectangular coordinate system the rule for a roadnetwork model construction composed of different roadsegments is represented as follows

(i) If and only if 119899 (119899 le 5) roads 1198771198981 1198771198985 intersectat an approximate point suppose that the road 1198771198981is defined by the line segment 1198771198981119909 perpendicularto 119909-axis so roads 1198771198982 and 1198771198985 adjacent to theroad 1198771198981 are represented as line segments 1198771198982119910 and1198771198985119910 intersected with the line segment 1198771198981119909 andperpendicular to 119910-axis and roads 1198771198983 and 1198771198984 notadjacent to road 1198771198981 are separately defined by theline segments 1198771198983119909 and 1198771198984119909 intersected with the linesegment119877119898119894119910 (1198771198982119910 or1198771198985119910) and perpendicular to119883For example there are five line segments intersectedat a point in Figure 2

(ii) If and only if three different roads119877119894119877119895 and119877119896 inter-sect at three points (as shown in Figure 3) supposethat the road 119877119894 is defined by the line segment 119877119894119909perpendicular to 119909-axis then the road 119877119895 is definedby the line segment 119877119895119910 intersected with the linesegment 119877119894119909 and perpendicular to 119910-axis and theroad 119877119896 is divided into two segments one is the linesegment 119877119896119909 intersected with the line segment 119877119894119909and perpendicular to 119909-axis and another is the linesegment119877119896119910 intersectedwith the line segment119877119895119910 andperpendicular to 119910-axis

The length of each line segment is defined as followsthe length of the line segment 119877119894119909 (Dist119877119894119909 = |1198841198942 minus 1198841198941|) isrepresented as the amount of line segments perpendicularto 119910-axis between two endpoints of 119877119894119909 (including twoendpoints) and the length of the line segment 119877119894119910 (Dist119877119894119910 =|1198831198942minus1198831198941|) is represented as the amount of line segments per-pendicular to 119909-axis between two endpoints of 119877119894119910 (includingtwo endpoints) But in Figure 3 the length of 119877119896 is differentfrom others The definitions for the length of 119877119896119909 and 119877119896119910 areboth limited in the region made up of roads 119877119894 119877119895 and 119877119896

Therefore as shown in Figure 4 our method transformsthe map into the road network model in a rectangularcoordinate systemOurmethod only deals withmain roads inthe map to clearly describe the process of building the model

4 The Definition of Driving Routes in119909-Axis and 119910-Axis

Suppose that the starting point of the vehicle route is 119860and the endpoint is 119861 the route composed of 119899 roads1198771 1198772 119877119899 from 119860 to 119861 is expressed as an ordered


HMM correction

Vehicle V1

Vehicle V2

Vehicle Vn

middot middot middot

Driving routeprediction

based on HMM

Entireroutes

Routerecommendation

Traffic conditionpreestimation

Vehicle Vi

A set ofOutput

Input

RS(Tk Roadi)

RouteT119896

Just-drivenroad segments


upcomingroutes

Figure 1 The architecture of route recommendation system

Rm1Rm2

Rm3

Rm4

Rm5

Rm1x

Rm2y

Rm3x Rm4x

Rm5y

Y

X0

Figure 2 Five roads intersect at a point

Ri

Rj

Rk

Rix

Rjy

Rkx

Rky

Y

X0

Figure 3 Three different roads intersect at three points

coordinate pointsrsquo sequence composed of 119899 minus 1 coordinatepoints

119860119899

997888rarr 119861 = 1198771119909 (1198771119910)

cap 1198772119910 (1198772119909) 119877(119899minus1)119910 (119877(119899minus1)119909) cap 119877119899119909 (119877119899119910)

(1)

where119860 is represented as the endpoint of the line segment1198771119909or 1198771119910 119861 is represented as the endpoint of the line segment119877119899119909 or 119877119899119910 and 119877(119894minus1)119909 cap119877119894119910 is represented as the intersectionpoint of the line segments 119877(119894minus1)119909 and 119877119894119910

For example the line connecting point 119860 (ie Hua-fuyuan) with point 119861 (ie Kangrsquoai Hospital) is a drivingroute in Figure 5 The vehicle has passed through 5 roadsincluding Fujian Road Zhongfu Road Heilongjiang RoadJinmao Street and Xufu Alley Suppose that 119860 is the starting

point and119861 is the endpoint then the route can be representedas follows based on Figure 4

Huafuyuan 5997888rarr Kangrsquoai Hospital

= (1 3) (1 4) (3 4) (3 1)

(2)

5 Driving Route Predictions Based on HMM

51 AMethod of Extending Training Set Based on119870-Means++It is necessary to train the HMM from driversrsquo past historyIn particular the larger the size of training examples is themore accurate theHMMfor path predictions is In view of thelimitation of given training examples the training set cannotcontain all of routes that drivers will take in the future Sothe paper proposes a method of extending training examplesbased on 119870-means++ [11] It could enlarge the training dataas much as possible based on given training examples

After analyzing the given training examples it is foundthat starting and endpoints of vehicle routes are distributedin residential commercial and work areas People usuallygo to work from residential areas in the morning and thengo back from work areas or they will first go to commercialareas and then go home Therefore it is believed that vehicleroutes are generally regular in some extent so that a path canbe regarded as two return paths In addition it is also foundthat when traffic reaches its peak a driver will generally avoidcongested roads and select a route with the shortest time tothe destination In other times drivers will select the shortestdistance to the destination to save costs For a beginningand end of a path it is able to generate two kinds of routesaccording to different times

Last it is not sure howmany clusters the coordinate pointset 119901 should be classified beforehand so the 119870-means++algorithm to automatically classify coordinate points into 119896clusters is exploited in the paper Here it should be pointedout that the distance of vehicle routes in the same cluster israther short so that people would not have to drive from onepoint to another It is not necessary to calculate vehicle routesfor the above case This assumption will be verified in theexperiment


Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd

Central RdXufu Alley

Sichuan RdJinmao St

Longpan Rd

Jianning Rd

0 1 2 3 4 5 6 7 80

1

2

3

4

5

6

Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd

Central Rd

Xufu Alley

Sichuan Rd

Jinmao St

Longpan Rd

Jianning Rd

X

Y

Figure 4 An example of the road network model construction

Figure 5 A path between points 119860 and 119861

The algorithm of extending training examples based on119870-means++ is as follows (see Algorithm 1)

(i) Initialize coordinate point sets 119901 and 1199011015840 and an

extending route set New119863 (Lines 01-02)(ii) Traverse a given training set 119863 and read all of

vehicle routesrsquo starting points (1199091198941 1199101198941) and endpoints(119909119894119899 119910119894119899) and then insert these coordinate points intothe set 119901 Filter repeated coordinates in the set 119901which could get the set 1199011015840 composed of differentstarting and endpoints (Lines 03ndash07)

(iii) Use the119870-means++ algorithm to classify 1199011015840 and thenacquire 119899 clusters 1198621 119862119894 119862119899 (Line 08)

(iv) Traverse each cluster119862119894 and then distinguish whetheror not two coordinate points belong to the samecluster 119862119894 If not use the function Best route(119888[119894][119896]119888[119895][119897]) to calculate routes between two coordinatepoints (Lines 09ndash13)

52 Parameter Definitions of a HMM for Route Predic-tions Since it is necessary to input a driverrsquos just-drivenpath represented by coordinate points into a HMM andthen output future entire paths coordinate pointsrsquo sequencecorresponding to the just-driven path can be regarded as

an observation sequence and the corresponding sequencecomposed of different route sets can be regarded as a hiddenstate sequence 119876 The next gives details on the process of theHMM construction by following training examples (shownin (3)) Note the number of training examples is much morethan following data in practice

Training Examples Consider

1199051 lt (1 3) (1 4) (3 4) (3 1) gt

1199052 lt (3 1) (3 4) (1 4) (1 3) gt

1199053 lt (0 3) (1 3) (1 5) (4 5) gt

1199054 lt (0 3) (0 0) (0 4) (4 1) gt

1199055 lt (2 0) (2 1) (3 1) (3 2) (4 2) gt

1199051 lt (1 3) (1 4) (3 4) (3 1) gt

(3)

In (3) assume that 1199051 1199052 are routesrsquo symbols in orderto distinguish different vehicle routes The observation set 119881includes the starting symbol (lt) the end symbol (gt) anddifferent coordinate points Each observation is defined by119901119894119895 where 119894 is the number of route 119905119894 in the training set and119895 is the number of coordinate points in each route 119905119894 Forexample the observation set of the above training example isltgt (1 3) (1 4) (3 4) (3 1) (0 3) (1 5) (4 5) (0 0) (0 4)(4 1) (2 0) (2 1) (3 2) (4 2) And an observation sequence119874 is an ordered sequence of symbols and coordinate pointsfrom the starting to the end For example the observationsequence of the route 1199051 is 11990111 rarr lt 11990112 rarr (1 3) 11990113 rarr(1 4) 11990114 rarr (3 4) 11990115 rarr (3 1) and 11990116 rarr gt

Besides the definition of hidden states is relatively morecomplex than observation states At first assume that eachhidden state is defined by 119902119894119895 where 119894 is the number of route119905119894 in the training set and 119895 is the number of coordinatepoints in each vehicle route 119905119894 The hidden state set 119878includes the symbol ∙ being produced from the observationslt gt and different routesrsquo symbol sets (eg 1199051 1199052 1199053 )corresponding to different coordinate points For examplehidden states being produced from the above observationsof the route 1199051 are separately 11990211 rarr ∙ 11990212 rarr 1199051 1199053


Input A training set119863Output The extending training set New119863(1) Coordinate Point Set 119901 1199011015840 = 120601(2) Extending route Set New119863 = 120601(3) foreach (route 119905119894 in119863)(4) Starting point 119860 = (1199091198941 1199101198941)(5) End point 119861 = (119909119894119899 119910119894119899)(6) Insert 119860 and 119861 into the set 119901(7) 119901

1015840 = Filter(119901)(8) Cluster Set 119862 = 119870-means++ (1199011015840)

lowast 119888 = 119888[1] 119888[2] 119888[119899] which is 119899 clusters altogether lowast(9) for (int 119894 = 0 119894 lt 119899 119894++)(10) for (int 119895 = 119894 + 1 119895 lt 119899 119895++)(11) for (int 119896 = 0 119896 lt 119888[119894]length 119896++)

lowast 119888[119894]length represents the number of coordinate points in the 119894th cluster lowast(12) for (int 119897 = 0 119897 lt 119888[119895]length 119897++)(13) Insert Best route(119888[119894][119896] 119888[119895][119897]) into New119863

lowast 119888[119894][119896] represents the 119896th coordinate point in the 119894th cluster lowast

Algorithm 1 New Track (a training set119863)

11990213 rarr 1199051 11990214 rarr 1199051 11990215 rarr 1199051 1199055 and 11990216 rarr ∙ Ahidden state sequence set is defined by QS storing hiddenstate sequences 119876 being produced from hidden states andeach vehicle route is directed Suppose that119860 119899997888rarr 119861 representsthat a vehicle passes through 119899 road segments from thestarting point 119860 to the endpoint 119861 but 119861 119899997888rarr 119860 representsthat a vehicle passes through the same road segments from119861 to 119860 Even though each observation state is same in thetwo opposite routes ordered coordinate pointsrsquo sequencesare completely opposite So a method is explored to calculatehidden states corresponding to each coordinate point next

The algorithm for hidden state determinations is asfollows (see Algorithm 2)

(i) Initialize a hidden state sequence set QS (Line 1)(ii) Obtain a beginning point119860 119894(1199091198941 1199101198941) and an endpoint

119861119894(119909119894119899 119910119894119899) from the vehicle route 119905119894 and a beginningpoint 119860119895 = (1199091198951 1199101198951) and an endpoint 119861119895 = (119909119895119899 119910119895119899)from the vehicle route 119905119895 then calculate 997888997888997888rarr119860 119894119861119894 = (119909119894119899 minus1199091198941 119910119894119899minus1199101198941) denoted by 119886119894 and

997888997888997888997888rarr119860119895119861119895 = (119909119895119899minus1199091198951 119910119895119899minus

1199101198951) denoted by 119886119895 (Lines 2ndash9)(iii) Compute the cosine value of intersection angle

between vectors 119886119894 and 119886119895 (Line 10)

cos ⟨ 119886119894 119886119895⟩ =

119886119894 sdot 119886119895

1003816100381610038161003816 1198861198941003816100381610038161003816 sdot10038161003816100381610038161003816119886119895

10038161003816100381610038161003816

= ((119909119894119899 minus 1199091198941) sdot (119910119894119899 minus 1199101198941)

+ (119909119895119899 minus 1199091198951) sdot (119910119895119899 minus 1199101198951))

sdot (radic(119909119894119899 minus 1199091198941)2+ (119910119894119899 minus 1199101198941)

2

sdotradic(119909119895119899 minus 1199091198951)2

+ (119910119895119899 minus 1199101198951)2

)

minus1

(4)

(iv) If 0 le cos⟨ 119886119894 119886119895⟩ le 1 traverse each coordinate pointin vehicle routes 119905119894 and 119905119895 and then judge whether ornot a coordinate point 119900119896

1

in 119905119894 is also included in 119905119895 Ifit is included insert a symbol 119905119895 into the correspond-ing location of the sequence 119876119894 (Lines 10ndash14) If minus1 ltcos⟨ 119886119894 119886119895⟩ lt 0 driving directions of the two routes areopposite although the routes include the same coordi-nate point For example if a vehicle is driving east ina route 119905119894 the possibility of passing through south orwestern roads in a route 119905119895 in our road networkmodelis low So the kind of hidden states will not be takeninto account And then insert a symbol ∙ and a symbol119905119894 into 119876119894 on the basis of the given 119876119894 (Lines 15ndash20)

(v) After calculating all of the hidden state sequenceinsert each hidden state sequence119876 into the sequenceset QS (Line 21)

53 Parameter Estimation of a HMM for Route PredictionsAfter determining observation states and corresponding hid-den states in theHMMfor route predictions ourmethod usesthe total training dataset Total119863 including the given trainingset119863 and the extending training set New119863 to estimatemodelparameters To reduce the negative impact on the HMM aweightedmethod is used to improve the process of estimatingHMM parameters In addition the problem of data sparse-ness also known as the zero-frequency problem arises in theprocess of building theHMM So ourmethod adopts the add-one (Laplace) [12] smoothing technique to deal with eventsthat do not occur in the total training set The process ofestimatingHMMparameters by a weightedmethod and add-one (Laplace) smoothing is described as follows

(i) The following equation is used for the initial proba-bility distribution

120587119894 =

Count (119904119863119894

) + 120582Count (119904New119863119894

)

sum119899

119895=1[Count (119904119863

119895

) + 120582Count (119904New119863119895

)]

(5)


Input A training set119863Output A hidden state sequence set QS(1) Hidden state sequence set QS = 120601(2) for (int 119894 = 1 119894 lt 119898 119894++)

lowast 119898 is the number of routes in119863 lowast(3) Starting point 119860 119894 = (1199091198941 1199101198941)(4) End point 119861119894 = (119909119894119899 119910119894119899)(5) Vector 119886119894 = (119909119894119899 minus 1199091198941 119910119894119899 minus 1199101198941)(6) for (int 119895 = 119894 + 1 119895 lt 119898 119895++)(7) Starting point 119860119895 = (1199091198951 1199101198951)(8) End point 119861119895 = (119909119895119899 119910119895119899)(9) Vector 119886119895 = (119909119895119899 minus 1199091198951 119910119895119899 minus 1199101198951)(10) if (0 le cos⟨ 119886119894 119886119895⟩ le 1)(11) foreach (Coordinate point 1199001198961 in 119905119894)(12) foreach (Coordinate point 1199001198962 in 119905119895)(13) If (119900

1198961= 1199001198962)

(14) Insert a symbol 119905119895 into 119876119894 corresponding to the coordinate point(15) else(16) foreach (Coordinate point 119900119895 in 119905119894)(17) If (119900119895 is a symbol ldquoltrdquo or ldquogtrdquo)(18) Insert a symbol ∙ into 119876

119894corresponding to the starting and end point

(19) else(20) Insert a symbol 119905119894 into 119876119894 corresponding to each coordinate point(21) Insert each hidden state sequence 119876 into the sequence set QS

Algorithm 2 Hidden State Sequence (a training set119863)

where 119899 is the number of hidden states (ie thetotal number of different vehicle routes) Count(119904119863

119894

)

and Count(119904New119863119894

) separately represent the numberof times the hidden state 119904119894 appears in the given andextending training sets and 120582 represents the weight(0 lt 120582 lt 1)

(ii) The following equation is used for the hidden statetransition matrix

119875 (119904119894 | 119904119894minus1)

=

Count (119904119863119894minus1

119904119863119894

) + 120582Count (119904New119863119894minus1

119904New119863119894

) + 1

Count (119904119863119894minus1

) + 120582Count (119904New119863119894minus1

) + 119898

(6)

where Count(119904119863119894minus1

119904119863119894

) and Count(119904New119863119894minus1

119904New119863119894

)

separately represent the number of times a hiddenstate 119904119894 followed 119904119894minus1 in the given and extendingtraining sets and119898 is the number of times the hiddenstate 119904119894 occurs in the total training set

(iii) The following equation is used for the confusionmatrix

119875 (V119895 | 119904119894)

=

Count (119904119863119894minus1

V119863119894

) + 120582Count (119904New119863119894minus1

VNew119863119894

) + 1

Count (119904119863119894

) + 120582Count (119904New119863119894

) + 119899

(7)


V119863119894


VNew119863119894

)

separately represent the number of times the hiddenstate 119904119894 accompanies the observation state V119895 in thegiven and extending training sets and 119899 is the numberof times the observation state V119895 occurs in the totaltraining set

As described above our method could build the HMMfor vehicle route predictions But drivers would like to choosedifferent vehicle routes from a starting point to an endpointduring different time of each day For example people hopeto reach the end during the rush hour (700sim900 AM and1700sim1900 PM) as quickly as possible and try their best toavoid congested roads But at other times people may choosethe shortest route to drive Therefore training examples canbe classified according to the time of day A group of trainingexamples is from 700sim900 AM and 1700sim1900 PM andanother is from other times Section 7 will test the impact onthe prediction accuracy with different training examples bybuilding different HMMs at different times

54 Driving Route Predictions The aim of this section is tointroduce how to predict upcoming routes based on just-driven road segments The solution to this problem is corre-sponding to aHMMdecodingwhich is to discover the hiddenstate sequence that was most likely to have produced a givenobservation sequence Here the Viterbi algorithm [13] is usedto find the best hidden state sequence composed of differentsymbols for an observation sequence (a given vehicle route)The process of a vehicle route prediction is shown in Figure 6


Input(1) A given HMM(2) An observation

sequence

Viterbialgorithm

A hidden state Routeprediction

OutputA set of upcomingvehicle routessequence

Figure 6 The process of driving route prediction

Input An observation sequence 119874Output A set 119877 of upcoming vehicle routesrsquo symbols(1) Ordered Observation Set 11986311198632 = 120601(2) Possible Route Set 119877 = 120601(3) Foreach (Observation 119901119894119895 in 119874)(4) if (119901119894119895 isin 119881)(5) lowast 119881 is a set of all of observations in the training set lowast(6) Insert 119901119894119895 into1198631(7) else(8) Insert 119901119894119895 into1198632(9) int119898 = length of1198631(10) int 119899 = length of1198632(11) if (119898 = 0)(12) 119877 = 120601(13) else if (119899 = 0)(14) 119877 = Viterbi Route (1199011198941 1199011198942 119901119894119896)(15) else if (119898 = 1 and1198631(1) = 1199011198941)(16) lowast 1198631(1) represents the first element in the set1198631 lowast(17) 119877 = Viterbi Route (1199011198941)(18) else if (1198632(1) = 119901119894119896)(19) Possible Routes (1199011198941 1199011198942 119901119894(119896minus1))(20) else if (1198632(1) = 1199011198941)(21) Possible Routes (1199011198942 119901119894119896)(22) else(23) Possible Routes (119901119894(119895+1) 119901119894119896)

Algorithm 3 Possible Routes (an observation sequence 119874)

Perhaps it will encounter some problems in the processof implementing Viterbi algorithm The total training setincluding the given and extending training examples is stillso limited that it could not fully contain all of possibleupcoming vehicle routes Assuming that the upcoming routedoes not occur in the total training set which means (1)part of coordinate points are new ones for training examplesand (2) each coordinate point has occurred in the totaltraining set a group from these coordinate points doesnot appear in the training examples For this case (1) theViterbi algorithm could not be directly used to compute thehidden state sequence For example in Figure 5 if a vehicleis on the current road segment represented by (4 4) and therepresentation of the corresponding just-driven route is 1199056 lt(0 3)(1 3)(1 4)(4 4) the Viterbi algorithm is not adoptedto find hidden state sequence for this observation sequenceAnd for case (2) even though the Viterbi algorithm canbe used each hidden state will not contain this new routersquossymbol For example if a new route is represented by 1199056 lt

(0 3)(1 3)(1 4)(3 4)(3 2) and all of these coordinate pointshave occurred in Figure 5 the symbol 1199056 of the upcomingvehicle route will not appear in each hidden state whichmeans people could not directly understand where the

vehicle will drive to Applied to these problems an algorithmfor vehicle route predictions is proposed as follows (seeAlgorithm 3)

(i) Suppose that 119874 = 1199011198941 1199011198942 119901119894119896 is an observationsequence composed of 119896 coordinate points after thevehicle has passed through 119896 roads then initializethree sets 1198631 1198632 and 119877 where 119877 represents aset of upcoming vehicle routesrsquo symbols 1198631 =

119901119894(1199091) 119901119894(119909

2) 119901119894(119909

119898) (1198631 isin 119881 as described above

119881 is a set of all of observations in the training set)1198632 = 119901119894(119910

1) 119901119894(119910

2) 119901119894(119910

119899) (1198632 notin 119881) and the

elements of 119874 are all in the set1198631 cup 1198632 (Lines 1-2)(ii) Traverse the observation sequence 119874 and determine

whether or not each coordinate point belongs to theset 119881 If a coordinate point belongs to 119881 then insertthe point into the set1198631 If not insert it into1198632 (Lines3ndash8)

(iii) Define that119898 is the number of elements in the set1198631and 119899 is the number of elements in the set 1198632 (Lines9-10)

(iv) If119898 = 0 the Viterbi algorithm is not used to find theupcoming routes and then 119877 = 120601 (Lines 11-12)


(1) Hidden state sequence 119876 = Viterbi(1198741015840)(2) int119898 = length of 119876(3) if (119898 = 1)(4) 119877 = 1198761(5) else(6) for (int 119894 = 2 119894 lt Num of 119876 119894++)(7) if (119877 cap 119876119894 = 120601)(8) 119877 = 119877 cap 119876119894(9) else(10) 119877 = 119876119894

Algorithm 4 Viterbi Route (an observation sequence 1198741015840)

(v) If 119899 = 0 theViterbi algorithm could be used to predictand then use a function Viterbi Route to acquire theroute set related to the upcoming routes most likelyThis set will be helpful for people to drive as much aspossible (Lines 13-14)

(vi) If the input observation sequence119874 has not appearedin the total training set before and part of coordinatepoints in119874 have also not appeared in119881 (ie1198632 = 120601)four cases should be discussed

(a) Suppose that 1198632 = 1199011198942 119901119894119896 then possibleroutesrsquo set could be calculated by the functionViterbi Route (1199011198941) (Lines 15ndash17)

(b) Suppose that 1198632 = 119901119894(1199101) 119901119894(119910

2) 119901119894119896 then

use the function recursion to predict with theobservation sequence composed of remainingcoordinate points 1199011198941 1199011198942 119901119894(119896minus1) (Lines 18-19)

(c) Suppose that 1198632 = 1199011198941 119901119894(1199102) 119901119894(119910

119899) then

use the function recursion to predict with theobservation sequence composed of remainingcoordinate points 1199011198942 1199011198943 119901119894119896 (Lines 20-21)

(d) In addition to the above cases suppose that1198632 = 119901119894(119910

1) 119901119894(119910

2) 119901119894(119910

119899) and 1199101 = 1 119910119899

= 119896 119898 = 1 then use the function recursionto predict with the observation sequence com-posed of remaining coordinate points 119901119894(119910

1)

119901119894(1199102) 119901119894(119910

119899) (Lines 22-23) For example the

input observation sequence is (0 3) (1 3) (1 4)(4 4) (4 5) where (4 4) notin 119881 then the resultof vehicle route prediction is the set of hiddenstates corresponding to the coordinate point(4 5)

The function Viterbi Route is described as follows (seeAlgorithm 4)

(i) Use Viterbi algorithm to calculate the hidden statesequence 119876 corresponding to the observationsequence 1198741015840 (Line 1)

(ii) Define that the number of elements in the hiddenstate sequence 119876 is119898 (Line 2)

(iii) If119898 = 1 a set 119877 of upcoming vehicle routesrsquo symbolsis the hidden state set 1198761 (Lines 3-4)

(iv) Calculate the intersection between 119877 and anotherhidden state set 119876119894 If this intersection exists 119877 =

119877 cap 119876119894 If not 119877 = 119876119894 (Lines 5ndash10)

For example if two hidden states are separately 11990211 rarr1199051 1199053 and 11990212 rarr 1199051 then 119877 = 1199051 1199053 cap 1199051 = 1199051 andthe most likely upcoming route is 1199051 If two hidden states areseparately 11990211 rarr 1199053 and 11990212 rarr 1199051 and 1199053 cap 1199051 = 120601then the most likely upcoming route is 1199053

6 Route Prediction Results

61 Experimental Platform Every vehicle should be equip-ped with a device for collecting vehicle route data And datacollectors use a mobile phone with software Map Plus Wemainly focus on one of functions path tracking to recorddown the path of driving It runs in the background whilesomeone could run other apps or lock the device at the sametime It also can export or send tracked paths as KML filesHowever continued use of GPS running in the backgroundcan dramatically decrease battery life of mobile phone Sothe experiment also needs an external large-capacity batteryto support the phone continuously In addition researchersinstall the software Google Earth on the computer to presenteach of collected vehicle routes

62 Data Collection A total of 20 volunteers are selected forthe purpose of collecting the experimental data In order tofacilitate the communication between volunteers and us allvolunteers are fromour university including 15 teachers and 5students A month later our researchers finally acquire a totalof 1052 paths where the number of different routes is 51 Thesame path is the journey that volunteers start from a point tothe end through the same road segments But in the processof the data collection there are some problems inevitably

(i) In tunnels underground parking and high-rise denseareas the phenomenon that part of paths are offsetfrom GPS noise will appear [14]

(ii) Volunteers forget to open the software for recordingroute data resulting in collecting route data unsuc-cessfully

(iii) Volunteers forget to turn off the software when theydrive to the end resulting in the path to be relativelyconcentrated in a small area

Once researchers come across the above problems whenchecking path data we will manually correct the GPS dataIn summary the experimental results can overcome theinfluence of GPS noise and human factor to ensure theaccuracy of the collected data

In the actual process of collecting the GPS data collectivedata do not only focus on the longitude and latitude but alsocombine the GPS data of the starting point the middle andthe end with road segments describing the route as a paththat is made up of the starting and endpoints and drivenstreets

63 Experimental Metric To evaluate the performance ofroute predictions based on HMM a metric to explore is the


correct prediction accuracy based on driven process Supposethat a vehicle has passed through 119894 roads the possible routeset 119877 after predicting based on HMM is 119877 = 1198771 1198772 119877119899So the definition of the prediction accuracy is as follows

119875119894 =sum119899

119896=1119863(119877119896 119862119877)

sum119899

119905=1Dist 1003816100381610038161003816119877119905

1003816100381610038161003816

times 100 (8)

where 119862119877 indicates an entirely upcoming route 119863(119877119896 119862119877)represents the number of duplicate road segments betweenone of possible vehicle routes in the set119877mdash119877119896 and the entirelyupcoming route and Dist|119877119905| represents the length of theroute 119877119905 that is the number of road segments

For example assume that the total training examples areshown in (3) and 1199051 is the upcoming vehicle route whichmeans 119862119877 is 1199051 from the starting point (1 3) to the end(3 1) When the vehicle has traveled through one road theobservation sequence 119874 is denoted by 119874 =lt (1 3) and thecorresponding hidden state sequence is 119876 = ∙ 1199051 1199053 So theduplicate between 1199051 and 1199051 1199053 separately is 119863(1198771 1198771) = 6119863(1198773 1198771) = 1 The length of routes 1198771 and 1198773 is separatelyDist|1198771| = 6 andDist|1198773| = 7 So when the vehicle has passedthrough the first point the prediction accuracy is as follows

1198751 =Repeat (1198771 1198771) + Repeat (1198773 1198771)

Dist 100381610038161003816100381611987711003816100381610038161003816 + Dist 10038161003816100381610038161198773

1003816100381610038161003816

times 100

=6 + 1

6 + 7times 100 = 5385

(9)


641 Training and Test Data In the experiment all ofcollected route examples are from the software Map Pluswhere each route is included in a KML file composed of aseries of GPS data Researchers check these data in a certaintime period through Google Earth According to previousdescription of the road networkmodel routes represented byGPS data points could be changed into ones represented bycoordinate points

Besides some extending training examples are intro-duced here These examples are extended from originalcollected data through a method to enlarge the training setbased on 119870-means++ described before Firstly raw trainingexamples composed of coordinate points have been enteredThen all of starting and endpoints can be divided into 5clusters based on 119870-means++ It is known that the distancebetween each coordinate point and the corresponding clus-tering center is on average 0314 km and the farthest distancebetween two points in a cluster is on average 0628 km Itcan illustrate that the distance between two places in a clusteris relatively short so most of people would not like to driveTherefore this is the reason that extending algorithmwas notused to calculate driving route in a cluster

Figure 7 displays the trip data overlaid on two mapsone of original different routes (a) and the other of originaland extending different routes (b) The number of extendingtraining examples is 13605 where the number of routesdifferent from original training examples is 13556

Finally the composition of test training examples isillustrated in detail To test the prediction accuracy of ourprediction algorithm ourmethod should acquire part of real-world vehicle route data Here the method applies a leave-one-out approach [4 15] meaning that part of route data areextracted from total training examples as test examples

Test Examples (i) It includes part of routes that have notappeared in the training examples So it can simulate real-world trip data to evaluate the prediction accuracy of ouralgorithm in actual applications

Test Examples (ii) All of the route examples have appeared inthe training examples It can evaluate the prediction accuracycompared to test examples (i) in order to illustrate a factthat the number of different routes in the training examplesshould be as much as possible

642 Prediction Accuracy Figure 8 shows the average cor-rect prediction rate of test examples (i) and test examples (ii)by percent of route completed and by current travel distancewith different weight values and also shows the comparisonof results between Jon Froehlichrsquos algorithm and our methodin these graphs ldquoPercent of trip completedrdquo is an intuitiveevaluation criterion and it is useful in evaluating how wellthe algorithm performed However it is difficult to achievein practice A vehicle navigation system can never be sure ofhow far along a route it is in terms of percentage completedwithout knowing the exact route of the trip from start-to-endmdashthis is what our prediction method is trying to predictInstead a much more practical input parameter is the triprsquoscurrent distance traveledmdashthat is how far the vehicle hastraveled since the trip began Furthermore it also shouldevaluate the weight value 120582 to impact HMM for driving routeprediction The algorithm separately set the threshold value120582 as 02 05 and 08

For test examples (i) Figure 8(a) shows that as expectedafter a vehicle has driven the first road segment little infor-mation is known about its path and the correct predictionrates of both algorithms are much lower After 35 ofthe trip has been completed the correct prediction rateof our algorithm increases to on average 4969 and JonFroehlichrsquos algorithm only increases to on average 2994after 50 completion the correct prediction rate of ouralgorithm moves to on average 6252 and Jon Froehlichrsquosalgorithmmoves to on average 3854 Figure 8(c) canmoreaccurately show the performance of our proposed algorithmfor driving route prediction in a real-world scenario Bythe end of the first mile the correct prediction rate of ouralgorithm jumps to 3193 accuracy and by the tenth milethis percentage increases to 6112 And the results of JonFroehlichrsquos algorithm are only between 23037 and 292 foreach mile traveled up to 20 miles

For test examples (ii) Figures 8(b) and 8(d) show thatthe correct prediction accuracy for both algorithms is onaverage higher than the test dataset (i) In Figure 8(b) thepercentage of our algorithm jumps to 9086 accuracy at thehalfway point but Jon Froehlichrsquos algorithm can increase tothis percentage only after 65 of the trip has been completed


(a) (b)

Figure 7 The trip data overlaid on two maps one of original data (a) and another of original data and extending data (b)

100908070605040302010

01009080706050403020100

Trip completed ()

Cor

rect

pre

dict

ion

()

(a) Correct prediction rate of all trips by percent of trip completed

Cor

rect

pre

dict

ion

()

100908070605040302010

01009080706050403020100

Trip completed ()

(b) Correct prediction rate of repeated trips by percent of trip completed

Cor

rect

pre

dict

ion

()

Current travel distance (km)0 2 4 6 8 10 12 14 16 18 20

100908070605040302010

0

Jon Froehlichrsquos algorithm120582 = 08

120582 = 05

120582 = 02

(c) Correct prediction rate of all trips by miles driven

Cor

rect

pre

dict

ion

()

100908070605040302010

0



120582 = 05

120582 = 02

(d) Correct prediction rate of repeated trips by miles driven

Figure 8 The performance of our prediction algorithm and Jon Froehlichrsquos algorithm

In Figure 8(d) by the end of first mile the correct predictionaccuracy is similar to Figure 8(c) but as the trip progressesthere is a significant jump in prediction accuracy By the endof 10 miles the percentage of our algorithm already increasesto 8387 but at this time Jon Froehlichrsquos algorithm onlyincreases to 63 As the vehicle has traveled up to 20 milesthe percentage of our algorithm can move to 9929

Figure 8 concludes that the accuracy for driving routepredictions increases as the number of observed road

segments increases This means that a longer sequence ofroad segments will be more helpful for our predictions Alsoboth of algorithms should take the driving direction intoaccount by the end of first road segment because the vehiclecould be heading toward either end of the current roadsegment and observing only one segment is not indicative ofa driverrsquos direction so that the correct prediction rate is nearlyzero Furthermore the prediction accuracy for repeated tripsis already on average much higher than for unknown trips


90

80

70

60

50

40

30

20

10

0Other time periods

Cor

rect

pre

dict

ion

()

Time of day

The average prediction accuracy by percent of route completedand by current travel distance with 120582 = 02

All tripsRepeated trips

700ndash900 AM and1700ndash1900 PM

Figure 9 Our algorithmrsquos sensitivity to time of day

It can demonstrate the necessity of extending the trainingexamples The probability that new routes occur will bereduced so that the prediction accuracy will be improved asmuch as possible At last the larger the threshold value ldquo120582rdquois the lower the correct prediction rate is In our opiniondriving routes are relatively regular but many route datafrom extending examples do not follow this rule Indeedit will disturb this rule to drop the prediction accuracy Onthe other hand we have to acquire these extending sampleswhich could improve the prediction accuracy as mentionedbefore Therefore we should keep balance meaning thatextending data not only reduces the impact on a driverrsquosregularity (a regular route is a path that a driver often takes)as much as possible but also keeps it in existence (in thetraining set) for training and improving the accuracy ofHMM It is similar to core thought of add-one (Laplace)smoothing for the problem of data sparsenessThis thresholdvalue is defined as 120582 = 001 in future applications

Figure 9 shows the results of prediction accuracy basedon different HMMs by the percent of trip completed and bycurrent travel distance depending on the time of day intotwo categories (i) 700sim900 AM and 1700sim1900 PM and(ii) other time periods Then HMMs are trained and testedaccording to classified test examples The plot shows that theprediction accuracy is not very sensitive to the time of dayso this is not an important factor to consider when makingdriving route predictions Froehlich and Krumm [4] alsofound a similar lack of sensitivity to both time of day andday of week for increasing prediction accuracy Above all it isnot necessary to classify training samples to acquire differentHMMs for route predictions according to the time of day

7 Conclusion

This paper firstly presents a driving route recommenda-tion system where the prediction module is the core ofrecommendation system thereby giving details on a method

to accurately predict a driverrsquos entire route very early in atripThen a road networkmodel was defined and normalizedeach of driving routes in the rectangular coordinate systemThemethod also builds HMMs tomake preparation for routeprediction using a method of training set extension based on119870-means++ and the add-one (Laplace) smoothing techniqueNext the paper introduces how to predict upcoming routes ina trip by HMMs and Viterbi algorithm Finally experimentalresults demonstrate the correction of our assumptions asmentioned before and also verify the effectiveness of ouralgorithm for routes predictions

As a direction of the future work the improvement willbe from two points (i) investigate to enhance the Laplacesmoothing technique to suit HMM for driving route predic-tions (ii) apply the statistics method to make Viterbi algo-rithm work with unknown coordinate points



Acknowledgments

The research is support by National Natural Science Foun-dation of China (nos 61170065 and 61003039) Peak ofSix Major Talent in Jiangsu Province (no 2010DZXX026)China Postdoctoral Science Foundation (no 2014M560440)Jiangsu Planned Projects for Postdoctoral Research Funds(no 1302055C) and Science amp Technology Innovation Fundfor higher education institutions of Jiangsu Province (noCXZZ11-0405)

References

[1] AHamilton BWaterson T Cherrett A Robinson and I SnellldquoThe evolution of urban traffic control changing policy andtechnologyrdquo Transportation Planning and Technology vol 36no 1 pp 24ndash43 2013

[2] A Karbassi andM Barth ldquoVehicle route prediction and time ofarrival estimation techniques for improved transportation sys-temmanagementrdquo in Proceedings of the IEEE Intelligent VehiclesSymposium pp 511ndash516 IEEE Columbus Ohio USA 2003

[3] J Krumm ldquoAmarkovmodel for driver turn predictionrdquo SAE SP2193(1) 2008

[4] J Froehlich and J Krumm ldquoRoute prediction from trip obser-vationsrdquo SAE SP 219353 SAE 2008

[5] R Simmons B Browning Y Zhang and V Sadekar ldquoLearningto predict driver route and destination intentrdquo in Proceedingsof the IEEE Intelligent Transportation Systems Conference (ITSCrsquo06) pp 127ndash132 IEEE September 2006

[6] D Tian Y Yuan J Zhou YWang G Lu andH Xia ldquoReal-timevehicle route guidance based on connected vehiclesrdquo inProceed-ings of the IEEE International Conference on Green Comput-ing and Communications and IEEE Internet of Things andIEEE Cyber Physical and Social Computing (GreenCom-iThings-CPSCom rsquo13) pp 1512ndash1517 Beijing China August 2013

[7] I Kaparias and M G H Bell ldquoA reliability-based dynamic re-routing algorithm for in-vehicle navigationrdquo in Proceedings ofthe 13th International IEEEConference on Intelligent Transporta-tion Systems (ITSC rsquo10) pp 974ndash979 IEEE September 2010


[8] J-W Lee C-C Lo S-P Tang M-F Horng and Y-H Kuo ldquoAhybrid traffic geographic routing with cooperative traffic infor-mation collection scheme in VANETrdquo in Proceedings of the 13thInternational Conference on Advanced Communication Tech-nology Smart Service Innovation through Mobile Interactivity(ICACT rsquo11) pp 1495ndash1501 IEEE February 2011

[9] I Leontiadis G Marfia D Mack G Pau C Mascolo and MGerla ldquoOn the effectiveness of an opportunistic traffic manage-ment system for vehicular networksrdquo IEEE Transactions onIntelligent Transportation Systems vol 12 no 4 pp 1537ndash15482011

[10] M H Kabir M N Alam and K K Sup ldquoDesigning anenhanced route guided navigation for intelligent vehicular sys-tem (ITS)rdquo in Proceedings of the 5th International Conference onUbiquitous and Future Networks (ICUFN rsquo13) pp 340ndash344 July2013

[11] XMa Y JWu YWang F Chen and J Liu ldquoMining smart carddata for transit ridersrsquo travel patternsrdquo Transportation ResearchPart C Emerging Technologies vol 36 pp 1ndash12 2013

[12] R Szalai and G Orosz ldquoDecomposing the dynamics of hetero-geneous delayed networks with applications to connected vehi-cle systemsrdquo Physical Review E vol 88 no 4 Article ID 0409022013

[13] N-S Pai H-J Kuang T-Y Chang Y-C Kuo and C-Y LaildquoImplementation of a tour guide robot system using RFID tech-nology and viterbi algorithm-based HMM for speech recogni-tionrdquo Mathematical Problems in Engineering vol 2014 ArticleID 262791 7 pages 2014

[14] B-F Wu Y-H Chen and P-C Huang ldquoA localization-assist-ance system using GPS and wireless sensor networks for pedes-trian navigationrdquo Journal of Convergence Information Technol-ogy vol 7 no 17 pp 146ndash155 2012

[15] J D Lees-Miller R E Wilson and S Box ldquoHidden markovmodels for vehicle tracking with bluetoothrdquo in Proceedings ofthe TRB 92nd Annual Meeting Compendium of Papers 2013

Research ArticleDetecting Traffic Anomalies in Urban Areas UsingTaxi GPS Data

Weiming Kuang Shi An and Huifu Jiang

School of Transportation Science and Engineering Harbin Institute of Technology Harbin 150090 China

Correspondence should be addressed to Huifu Jiang jianghuifu1987outlookcom

Received 21 November 2014 Revised 26 January 2015 Accepted 22 April 2015

Academic Editor Chi-Chun Lo

Copyright copy 2015 Weiming Kuang et alThis is an open access article distributed under theCreativeCommonsAttribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Large-scale GPS data contain hidden information and provide us with the opportunity to discover knowledge that may be usefulfor transportation systems using advanced data mining techniques In major metropolitan cities many taxicabs are equipped withGPS devices Because taxies operate continuously for nearly 24 hours per day they can be used as reliable sensors for the perceivedtraffic state In this paper the entire city was divided into subregions by roads and taxi GPS data were transformed into trafficflow data to build a traffic flow matrix In addition a highly efficient anomaly detection method was proposed based on wavelettransform and PCA (principal component analysis) for detecting anomalous traffic events in urban regions The traffic anomaly isconsidered to occur in a subregion when the values of the corresponding indicators deviate significantly from the expected valuesThis method was evaluated using a GPS dataset that was generated bymore than 15000 taxies over a period of half a year in HarbinChina The results show that this detection method is effective and efficient

1 Introduction

Traffic anomalies widely exist in urban traffic networks andnegatively effect traffic efficiency travel time and air pollu-tion [1] The traffic flow in a road network is abnormal whentraffic accidents traffic congestion and large gatherings andevents such as construction occur [2] Thus the detectionof traffic anomalies is important for traffic managementand has become important in transportation research [3]Fortunately most taxies in cities in China are equipped withGPS devices [2] Because taxies can use road networks widelyover long periods their trajectories can reflect the trafficcondition in the road network [4] In other words taxies canbe observed as ldquoflowing detectorsrdquo in the urban road networkThus the difficulty of collecting data is reduced so that peoplecan improve the detection of anomalies with a large volumeof data

Several data mining methods have been proposed toachieve the goal of detecting anomalies by using GPS dataMost previous studies can be divided into two categories (1)studies on taxi GPS trajectory anomalies and (2) studies ontraffic anomalies In the first category most studies focus on

how to observe a small number of drivers with travelling tra-jectories that are different from the popular choices of otherdrivers [5] Some of these studies can be used to detect fraud-ulent taxi driving behavior to monitor the behavior of taxidrivers [6ndash8] Others have paid more attention to hijackedtaxi driving behavior which can protect taxi drivers andpassengers from assaultive injury [9] With the developmentof vehicle navigation technology new interest in trajectoryanomaly research has occurred which can be integrated withnavigation to provide dynamic routes for drivers or travelers[10ndash13] In addition this research can provide accurate real-time advisor routes compared with navigation based on statictraffic information The purpose of the second category isdifferent from the above studies In the second categorydetection algorithms and optimization methods have beenused to detect anomalies and piece them together to explorethe root causes of anomalies [14 15] In addition some othermethods were proposed for monitoring large-area traffic [1617] and determining the defects of existing traffic planning[18]The differences between these two categories include thefollowing aspects First the comparison between trajectories



in the anomalous trajectory process always focuses on a smallnumber of trajectories and the remaining normal trajectoriesat the same location during a certain period Second thedetection of traffic anomalies is used to detect a large numberof taxies with anomalous behaviors and detect potentialevents with time

This research belongs to the traffic anomaly detectionsome relevant works are those researching anomaly detectionwith GPS data [14 19 20] and some others use social mediadata as the source of mobility data to detect anomalies [2122] Most of these methods can be grouped into four cat-egories distance-based cluster-based classification-basedand statistics-based categories [23 24] In this paper theresearch focuses on taxi GPS data and the detection methodcan be classified as statistics-based According to an analysisof the existing literatures most studies have only consideredtraffic volume velocity and other visualized parameters andhave not considered the spatial information hidden in thetraffic flow [25] Moreover most existing methods are simplemethods based on single detection methods [17 23ndash25] ormodified versions of traditional outlier detection methods[14] These methods can easily detect long-term anomaliesbut lose many short-term anomalies which can continue fora short period thus the focus of this study is to improve thesensitivity of detectionmethods Somemethods for detectinganomalies in computer networks or financial time series usethe wavelet transform method to improve the performanceof detecting rapid anomalous changes [26 27] This idea canbe introduced into this research to achieve the same goalbecause the road network is similar to the computer networkNext a traffic anomalies detection method was proposedwhich can be distinguished in two ways First this methodcombines the wavelet transform method and PCA to detecttraffic anomalies due to low or high rates of change in trafficflowTherefore thismethod canmore effectively detect trafficanomalies than other detection methods that only use PCA[14] Further this method can provide information regardingthe spatial distribution of traffic flows The advantage of thismethod is identifying the rootswhile detecting the anomalieswhich reduces the blindness of traffic guidance

The organizational structure of this paper is organizedas follows In Section 2 the GPS data transformation andthe anomalies detecting method are described in detail InSection 3 case study is conducted based on taxi GPS dataof Harbin and the effectiveness and performance of theproposed method are analyzed at the same time Finally inSection 4 the conclusions from this research are summarized

2 Material and Methods

Traffic anomalies always occur in regions with large trafficvolume or high road network densities and deviate due tochanges in external conditions when compared with theperformance of normal traffic Many factors can result intraffic anomalies including traffic accidents special trafficcontrols large gatherings demonstrations and natural dis-asters [1] These causes may lead to a wide range of traffic

Figure 1 Network-based urban area segmentation

changes and further produce anomalous traffic flow patternsFurthermore traffic anomaly levels can be serious because oftraffic flow propagation

21 Road Network Traffic and Traffic Flow Matrix

211 Road Network Traffic In the taxi GPS data each taxitrajectory consists of a sequence of points with ID num-ber latitude longitude vehicle state (passengeremptyno-service) and timestamp information Taxi drivers need tostop their vehicles to pick up or drop off passengers (referredto as a vehicle state transition) thus each trajectory canbe divided into several end-to-end subtrajectories that aredefined as ldquotriprdquo in this paper Because three types of vehiclestate are used the trips can be considered as ldquopassengerrdquo tripsldquoemptyrdquo trips and ldquono-servicerdquo trips

Although three types of vehicle state are used the ldquono-servicerdquo GPS points will be merged to one point in the map-matching process which can be ignored in this researchOnly two classes of the trips were investigated one is theldquopassengerrdquo trip and the other is the ldquoemptyrdquo trip Each triprepresents the behavioral characteristics of traveling from anorigin point 119874 to a destination point 119863 However any twotrips will not have the same origin point or destination point(spatial dimension) in real life Consequently road networktraffic is hidden among different trips and it is difficult todetect traffic anomaliesTherefore the transport networkwassimplified and a novel network traffic model was proposedfor in-depth analysis and reducing complexity Urban areaswere segmented into subregions by road networks [28] Asdemonstrated in Figure 1 each subregion is surrounded by acertain level of road and any two adjacent subregions do notoverlap in space This model can provide more natural andsemantic segmentation of urban spaces Next a traffic modelwas constructed based on urban segmentation In this modelthe vehicles mobility in the subregion was ignored and allsubregions were abstracted into nodesThe road network wasmodeled as a directed graph 119866 = (119873 119871) where 119873 is a setof nodes (subregions) and 119871 is a set of links that connecttwo adjacent subregions A link can represent the mobility of


Table 1 Virtual OD nodes pairs

Origin virtual node Destination virtual node1198811198731

1198811198732

1198811198733

1198811198734

1198811198731

(1198811198731 1198811198731) (119881119873

1 1198811198732) (119881119873

1 1198811198733) (119881119873

1 1198811198734)

1198811198732

(1198811198732 1198811198731) (119881119873

2 1198811198732) (119881119873

2 1198811198733) (119881119873

2 1198811198734)

1198811198733

(1198811198733 1198811198731) (119881119873

3 1198811198732) (119881119873

3 1198811198733) (119881119873

3 1198811198734)

1198811198734

(1198811198734 1198811198731) (119881119873

4 1198811198732) (119881119873

4 1198811198733) (119881119873

4 1198811198734)

vehicles between two adjacent subregions Meanwhile ldquotriprdquoand ldquopathrdquo must be redefined based on this new model

Definition 1 (trip) A trip tr is a time sequence consistingof subregions with timestamp and can be transformed intoa time sequence of nodes that can represent subregions in themodel (ie tr ⟨119873

1 1199051⟩ rarr ⟨119873

2 1199052⟩ rarr sdot sdot sdot rarr ⟨119873

119899 119905119899⟩)

Definition 2 (path) A path 119875 is a sequence of nodes withouttemporal information (ie tr 119873

1rarr 119873

2rarr sdot sdot sdot rarr 119873

119899)

A path can represent the common spatial trajectory of sometrips that have the same node sequences when the timestampis ignored

Definition 3 (trajectory) A trajectory 119879 is a sequence ofconnected trips (ie 119879 = tr

1rarr tr2rarr sdot sdot sdot rarr tr

119899) where

tr(119896+1)

sdot 119904 = tr119896sdot 119890 (1 le 119896 lt 119899) tr

(119896+1)sdot 119904 is the start node of

tr(119896+1)

and tr119896sdot 119890 is the end node of tr

119896

This road network traffic model can represent the spatialmobility characteristics of flows from the origin to destina-tion nodes Thus they not only flow within different nodesand links in the road network but also tell us how traffic flowsfrom origin nodes to destination nodes The road networktraffic is used to obtain the sizes of the OD traffic flows Allof the traffic in the network will flow from origin nodes andacross some different intermediate nodes and links beforereaching the destination nodesThismethod is useful becauseall of the network topology information can be expressedas shown in Figure 2 In the logical topology layer eachnode can be observed as an origindestination node andthe link between two nodes represents the traffic flow fromthe origin node to the destination node However when thelogical topology layer is mapped to the physical topologylayer each path of the logical topology layer is divided intoseveral different sequences of links as defined inDefinition 2This method can help us extract the traffic information fromtraffic flow data However in this research the aim is not onlyto detect which OD nodes pairs have anomalous traffic butalso to identify which trips between the OD nodes pairs areanomalous Further two concepts called ldquovirtual noderdquo andldquovirtual OD nodes pairrdquo are defined as follows

Definition 4 (virtual node) Virtual node is an imaginarynode Each node in this road network has at least one virtualnode and the virtual nodes have the same spatial-temporalcharacteristics as shown in Figure 2

Definition 5 (virtual OD nodes pair) The virtual OD nodespair is composed of virtual nodes with each virtual OD nodepair possessing traffic flow across a unique path Only theorigindestination nodes of the path can be represented by thevirtual node and the intermediate nodesmust be real VirtualOD node pairs can help us build different paths between thesame OD node pairs (ie 119875 = 119881119873

1rarr 119873

2rarr sdot sdot sdot rarr

119873119896minus1

rarr 119881119873119896 119896 = 1 2 where 119875 is a path and 119881119873

1

and119881119873119896are origin virtual node and destination virtual node

resp) As shown in Figure 2 there are four virtual OD nodepair paths (virtual node 3 rarr virtual node 1)The number of avirtual OD nodes pair is equal to the number of the path thatconnects the OD nodes

Next virtual OD node pairs were built according tothe logical topology layer as shown in Table 1 Based onthe information shown in Table 1 one node can connectwith multiple nodes and those multiple nodes can have thesame destination node Previously the network traffic featurewas formulated and the traffic model can hold the spatialcorrelation of traffic flows the network wide traffic is a timesequencemodel and the time and frequency properties of thetraffic can be held well In the next step a transform domainanalysis was conducted for the road network traffic to detecttraffic flow anomalies

212 Index Building An index structure was created foranomaly detection process Each OD node pair can haveseveral paths that can connect the OD nodes (virtual ODnodes) However the research goal is to determine whichpaths of the OD node pairs are anomalous Thus an indexstructure was built which is an offline index structurebetween the path and links that can connect the nodesvirtualnodes For example in Figure 3(a) the points represent thenodesvirtual nodes the solid directed lines represent thelinks and the dashed lines represent the paths between theOD nodes pairs This index method is offline but can beupdated to be online when new data are received as shownin Figure 3(b)

213 Traffic Flow Matrix The traffic anomalies detectingmethod based on multiscale PCA (MSPCA) in this paperuses the traffic flowsmatrix as a data sourceThus the relateddefinitions of the traffic matrix are presented as follows

Definition 6 (traffic flow matrix) A traffic flow matrix is thetraffic demand of all the virtual OD nodes pairs in a road


Subregion 1

Subregion 2

Subregion 3

Subregion 4

Node 1Node 4

Node 2Node 3

Virtual node 4

Virtual node 2Virtual node 3

Virtual node 1

Virtual node 4

Virtual node 2Virtual Node 3

Virtual node 1

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Physical topology

Logical topology

Figure 2 The road network model used for detecting network traffic anomalies

Link 2

Link 5

Link 1

Path 1 Path 2

Link 3

Link 4

Path 3 Path 4

(a) Logical topology

Link 1

Link 2 Link 3 Link 4

Link 5

Path 1

Path 2

Path 3

Path 4

Path 1Link 1

Link 3

Link 4

Path 2


Path 3Link 2

Link 3

Link 4 Path 2

Link 3Link 2

Path 3 Path 4Path 1 Path 2

Path 1 Path 3

Path 4

Link 4

Path 2

(b) Index

Figure 3 Example of the index

network The traffic flow matrix can be further classified asan NtN (node-to-node) traffic flow matrix

Definition 7 (NtN traffic flow matrix) If the network has119899 nodes and the traffic flow of any path can be measuredconstantly over a certain time interval then the measuredvalue can be created as a 119879 times 119908 matrix to represent a timesequence of the measured traffic flow Here 119879 is the numberof measured cycles and 119908 is the number of traffic flowmeasurements thus119908 = 119899 times 119899 Row 119905 is a vector of trafficflowvalue which ismeasured in the 119905 cycle and can be representedby 119909119905 The column 119895 is the time sequence of the traffic flow

value of 119895 virtual OD node pairs In addition 119909119905119895represents

the traffic flow of the 119895 virtual OD node pairs during the 119905cycle

[[[[[[[[

[

11990911

11990912

sdot sdot sdot 1199091119908minus1

1199091119908

11990921

11990922


1199092119908

sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot

119909119879minus11

119909119879minus12

sdot sdot sdot 119909119879minus1119908minus1

119909119879minus1119908

1199091198791

1199091198792


119909119879119908

]]]]]]]]

]

(1)


22 Traffic Anomaly Detection Method

221 Traffic Anomaly Detection Process The detection oftraffic anomalies from a wide traffic network can be obtainedby developing a method that can determine anomaloussubregions in a network to provide effective informationfor transportation researchers and managers for improvingtransportation planning and dealing with emergencies Gen-erally this problem can be described by considering howto capture the anomalous subregions whose characteristicvalues significantly deviate from normal values To achievethis goal a novel computing process was designed as shownin Figure 4 In this process the physical topology layer istransformed according to the structure of the real networkThen the logical topology layer can be derived and theOD nodes pairs and virtual OD nodes pairs are establishedsimultaneously Furthermore the traffic of the paths betweenthe virtual OD nodes pairs is extracted with logical topologyinformation while using the wavelet transform method andPCA to prove the spatial and temporal relationships Basedon the multiscale modeling ability of the wavelet transformand the dimensionality reduction ability of PCA the networktraffic anomalies detection method can be constructed basedon multiscale PCA with Shewhart and EWMA control chartresidual analyses Finally a judgment method is proposed fordetecting the anomalous location

222 Traffic Anomalies Detecting Method Based on MSPCAIn this section the space-time relativity of the traffic flowmatrix was used to model the ability of the wavelet transformand the dimensionality reduction of PCA to transform thetraffic flow of the traffic flow matrix Next anomalies weredetected using two types of residual flow analysis The timecomplexity analysis will be discussed at the end of thissection

Normal traffic flow modeling can be met by usingthe MSPCA which can combine the abilities of wavelettransform to extract deterministic characteristics with theability of PCA to extract the common patterns of multiplevariables Normal traffic flowmodeling based onMSPCA canbe divided into the four following steps

Step 1 The first step is the wavelet decomposition of thetraffic flow matrix First the traffic flow matrix 119883 willundergo multiscale decomposition through an orthonormalwavelet transform [29] Next the wavelet coefficient matrix119885119871 119884119898(119898 = 1 119871) can be obtained on every scale Then

theMADmethod [30] is used to filter thewavelet coefficientsFinally the following filtered wavelet coefficient matrix isobtained

119885119871 119884119898

(119898 = 1 119871) (2)

Step 2 The second step is principal component analysis andrefactoring of the wavelet coefficientmatrix First the waveletcoefficient matrix 119885

119871 119884119898(119898 = 1 119871) in every scale is

analyzed using PCA Next the number of nodes is selectedaccording to the scree plot method [31] Finally the waveletcoefficient matrix 119885

119871 119898(119898 = 1 119871) is reconstructed

Step 3 The third step is reconstructing the traffic flowmatrixusing the invert wavelet transform 119882

119879according to thewavelet coefficient matrix 119885

119871 119898(119898 = 1 119871) at all scales

Step 4 The fourth step is principal component analysis andrefactoring of the traffic flowmatrixThismethod is similar tothat of Step 2 and the traffic flowmatrix can be reconstructeddenoted by119883

After the normal traffic flow was modeled several resid-ual traffic flows were determined including two componentsnoise and anomalous traffic These flows mainly resultedfrom errors of the traffic flow model and traffic anomaliesrespectivelyThe squared prediction errorwas used to analyzethe residual traffic flows

SPE119894=

119882

sum

119895=1

(119909119894119895minus 119909119894119895)2

(3)

where 119909119894119895is the element in the traffic flow matrix119883 and119882 is

the number of links in the networkThen two types of control chart methods were used to

analyze the residual traffic flows Shewhart and EWMA [32]The Shewhart control chart method can detect rapid changesin traffic flow but its detection speed is slow for detectinganomalous traffic flows which change slowly However theEWMA control chart method can detect anomalous trafficflows that have a long duration but change slowlyShewhart Control Chart MethodThe Shewhart control chartmethod directly detects the time sequence of the squaredprediction error and defines 1205852

120572as the threshold for the

squared prediction error at the 1 minus 120572 confidence level Astatistical test known as the 119876-statistic [31] is used to test theresidual traffic flows as follows

1205852

120572= 1206011

[[

[

119888120572radic21206012ℎ2

0

1206011

+ 1 +1206012ℎ0(ℎ0minus 1)

1206012

1

]]

]

1ℎ0

(4)

where ℎ0= 1 minus 2120601

1120601331206012

2 120601119894= sum119882

119895=119903+1120582119894

119895 119894 = 1 2 3 120582

119895is

the variance which can be obtained by projecting the trafficflow matrix to the 119895th principal component 119888

120572is the 1 minus 120572

percentile in the standardized normal distribution and 119903 isthe intrinsic dimensionality of the residual traffic flows dataIf the value of the squared prediction error is not less than thethreshold value 1205852

120572 an anomaly will appear

According to the 119876-statistic the multivariate Gaussiandistribution follows the assumption of derivation The 119876-statistic will display few changes even when the distributionof the original data differs from the Gaussian distribution[31] Thus the 119876-statistic can provide prospective results inpractice without examining traffic flows data for adaptionassumptions due to its robustnessEWMA Control Chart Method The EWMA control chartmethod can be used to predict the value of the next momentin the time sequence according to historical data The pre-dicted value of residual traffic flow at time 119905 can be recorded


Transform

Physical topology

Logical topology

Taxi GPSdata

Traffic flowdata

Segmentedroad network Wavelet

transformPCA

Shewhart controlchart method

EWMA controlchart method

Anomaloustraffic flows

Judge

Anomalousposition

Figure 4 Traffic anomalies detection process

as119876119905 and the actual value of the residual traffic flow at 119905 is119876

119905

Thus

119876119905+1= 120573119876119905+ (1 minus 120573)119876

119905 (5)

where 0 le 120573 le 1 is the weight of the historical dataThe absolute value of the difference between the actual andpredicted values |119876

119905minus119876119905| is obtained and the threshold value

of EWMA can be defined as follows

120595 = 120583119904+ 119871 times 120590

119904radic

120573

(2 minus 120573) 119879 (6)

where 120583119904is the mean value of |119876

119905minus119876119905| 120590119904is the mean square

error 119871 is a constant and119879 is the length of the time sequenceThus if |119876

119905minus 119876119905| ge 120595 an anomaly will appear

The computational complexity of the proposedmethod is119874(1198791199012+ 119879119901) which mainly contains the wavelet transform

and PCA processCurrently the paths which have traffic anomalies can be

detected However the research goal is to determine whichlinks between the adjacent regions are anomalousThereforeanother method was designed to locate anomalous linksbased on the distribution of traffic flow in the next section

223 Anomalous Position Locating According to the analysisresults the paths of OD node pairs may have different trafficflow values at the same time However determining whichpaths are anomalous is not the purpose of this researchThe anomalous position should be located to provide usefuland clear information for transportation researchers andmanagers The proposed method is different from othermethods which detect the anomalous road segment firstand then infer the root cause of the traffic anomalies in theroad network Here the paths with traffic anomalies can bedetected and the anomalous position locating process wasbuilt as follows First the trips were connected with thepaths that have traffic anomalies so that all links belongingto an anomalous path can be identified Next all links areassumed as potential anomalous links and stored into ananomalous pool Next the existing identification method isused to determine whether traffic anomalies exist on theselinks based on their historical data this process ends until all

of the links are tested Finally the links that are not anomalousare deleted and the other links are kept in the anomalous pool

Links do not exist in the physical worldThus anomalouslinks need to be transformed into anomalous subregionsBased on the experience the subregions that are connectedby anomalous links will have the greatest probability of beinganomalous Thus all of these subregions should be searchedand considered as anomalous subregions The traffic flowbetween them is anomalous So far the process of trafficanomalies detection has been completely presented

3 Results and Discussions

31 The Road Network and Data Preparation

311 Road Network The road networks of Harbin wereconsidered as the basic road networks and the statisticalinformation is shown in Table 2 To obtain a higher detectionprecisionminor roads andmajor roads were used to segmentthe urban area as shown in Figure 5 (the green lines and bluelines are minor roads and major roads resp) Consequentlythe area of the subregions became smaller so that the trafficanomalies can be located more accurately Thus the numberof subregions significantly increases relative to the numbershown in Figure 1

312 Mobility Data The taxi GPS data were used as mobilitydata as shown in Table 2 Approximately 23 of the dailyroad traffic in Harbin is generated by taxies Thus taxitraffic can indicate the dynamics of all traffic Although themobility data were collected from taxies it can be believedthat the proposed method is general enough to use otherdata sources which can reflect the characteristics of mobilityon the road network such as the public transit GPS dataAll of these data require preprocessing to remove erroneousdata and eliminate positioning deviations by map-matchingtechnology

32 Evaluation Approach In the numerical experiment thetraffic anomalies reported during the half-year period wereused as real data to evaluate the detecting effectivenessand performance of this approach In practice continuousexecution is unrealistic due to the need for large amounts of


(a) 7ndash9 AM reported incidents (b) 4ndash6 PM reported incidents

(c) 7ndash9 AM baseline 1 results (d) 4ndash6 PM baseline 1 results

(e) 7ndash9 AM baseline 2 results (f) 4ndash6 PM baseline 2 results

(g) 7ndash9 AM proposed method results (h) 4ndash6 PM proposed method results

Figure 5 Reported traffic anomalies and detection results

computation thus time discretization was used to overcomethis fault The time interval of algorithm execution is 15minutes It means the detection method was executed every15 minutes with the data collected during the latest period ascurrent data All of the previous data were stored as historicaldata in the database and used for experimental calculationsIn addition the length of the time interval can be determinedbased on the actual demand (it is a tradeoff process readerscan refer to Ziebart et al [11])

321 Measurement In the process of evaluating the effec-tiveness of the proposed traffic anomalies detection methodtraffic anomaly reports were used as a subset of real trafficanomalies because not all traffic anomalies can be recordedin reports The evaluation method consists of comparing thedetection results with the reports to determine howmany realtraffic anomalies can be detected Thus the 119877 parameter wasdefined to measure the accuracy which can be expressed as119877 = 119862

119889119862119903 where 119862

119889is the number of reported anomalies


Table 2 Dataset statistics

Data duration MarndashAug 2012

GPS data

Taxies 15210Effective days 74

Trips 21510880Avg sampling interval 60 s

Road network Road grade Major and minor roadsSubregions 387

Reports Avg reports per day 28

that can be detected using the proposedmethod and119862119903is the

number of anomalies in the reports This parameter is nota precision measurement because a traffic anomalies reportmay not provide a complete set of all real traffic anomaliesIt is possible that some traffic anomalies can be detected byusing the proposedmethod but should not be recorded in thereport as shown in Figure 5

322 Baselines The accuracy of the proposed methodshould be evaluated in this process Two anomalous trafficdetection methods were used as baselines a method basedon the likelihood ratio test statistic (LRT) [17] and a modifiedversion of PCA [14] The ideas used in these two methodsare similar to ours thus these methods were applied to thematrixes of all subregions to find out the subregions whichhave an anomalous number of taxies based on our segmen-tation Next the accuracy can be obtained by comparing theresults of the three methods

33 Numerical Experiments

331 Effectiveness To accurately evaluate the proposedmethod two ldquopeak-hourrdquo time intervals on 1152012 werechosen as study period which are presented in Figure 5 (thered regions of all eight figures indicate the anomalies) Figures5(a) and 5(b) show the anomalies that were reported duringthese two time intervals Figures 5(c) and 5(d) show theanomalies that were detected by using baseline 1 method (themethod based on LRT) and Figures 5(e) and 5(f) show theanomalies that were detected by using baseline 2method (themodified version of PCA) In addition Figures 5(g) and 5(h)show the detection results of the proposed method

According to Figure 5 the proposed method detectedmore traffic anomalies than the baseline methods duringeach time interval From 7 AM to 9 AM baseline 1 methodand the proposed method detected all anomalies in thereport However baseline 2 method only detected 75 of theanomalies In addition the results show that the proposedmethod detected 2sim3 more anomalies (which could bepotential anomalies) than the baseline methods From 4PM to 6 PM the proposed method can detect 10 reportedanomalies However baseline 1 and 2 methods resulted in 8and 9 reported anomalies respectively Thus the proposedmethod can detect 9091 of all reported anomalies in thisspecial time interval which is 1818 more than the value of

baseline 1 method and 909 more than the value of baseline2 method In the experiments of different time intervals on1152012 the average 119877 value of the proposed method is8237 but the value of baseline 1 method is only 6374and the value of baseline 2 method is 7270 When theexperiment was extended to another 73 effective days fromMarch to August as shown in Table 3 the average 119877 valueof the proposed method is 7462 the value of baseline 1method is 5633 and the value of baseline 2 method is6329This phenomenon indicates that the detection rate ofthe proposedmethod improved by 3247 and 1790 relativeto baseline 1 and baseline 2methods respectively In additionaccording to the 119877 value of each day the proposed methodcan detect more reported anomalies than the baselinesThusit can be concluded that the proposed method is significantlybetter than the baseline methods

To further illustrate the feasibility and superiority ofthe proposed method an anomalous subregion was chosenbetween 730 AM and 930 AM In this case three anomalouspaths can be observed in the subregion (their traffic flowis shown in Figure 6) Thus the path that causes trafficis obvious and the transportation managers can guide thetraffic to the regions that have less traffic pressure

According to Figure 6(a) the overall traffic flow did notdiffer much from the regular overall traffic flow between 700AM and 745 AM However between 745 AM and 830 AMa significant difference was observed between the two curvesBy comparing Figures 6(b) and 6(c) this traffic anomalyresulting from the traffic flow of path A can be observedobviously According to Figure 6(d) the percentages of thetraffic flow in paths B and C declined between 745 AM and830 AM because some taxi drivers changed their routes toavoid this anomalous region After this period the trafficflow gradually returned to the normal status as shownin Figure 6(a) Consequently in the directions with morepotential capacity for sharing more traffic flows such as pathB in Figures 6(c) and 6(d) the traffic flow and percentages alldecreased during the anomalous interval thus a portion ofthe traffic flow can be guided to this direction to reduce thetraffic pressure of anomalous region

332 Performance In the experiments the hardwaresoft-ware configuration and average processing time for anomalydetection are shown in Tables 4 and 5 respectively Theurban area was segmented into a number of subregions inthe first step and the following study was affected by thesegmentation resultsThe computing times for different stepsare related to the numbers of subregionsThus the computingtimes will be significantly different when the urban area issegmented according to different levels of roads Specificallythe computing time will increase as the road level decreasesas shown in Figure 7

34 Case Study In this section two cases were used tofurther evaluate the detection method In the first case ananomalous region was detected and reported In anothercase the detected anomalous region does not exist in thereport these two cases are shown in Figures 8 and 9


Table 3 R values of the detection results

Number Date 119877 value of each dayBaseline 1 method Baseline 2 method Proposed method

1 432012 5927 6297 83172 632012 6418 6452 75863 732012 5344 7020 8849

32 1152012 6374 7270 8237

74 3182012 4728 7737 7888Average 119877 value 5633 6329 7462

050

100150200250300350400450500

Traffi

c flow

Flow in regularFlow in anomaly

t

700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900

(a) Traffic flow comparison

t

0

20

40

60

80

100

120

140

Traffi

c flow

Path A in regularPath B in regularPath C in regular

700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900

(b) Regular traffic flow of paths

t

0

50

100

150

200

250

300

350

Traffi

c flow

Path A in anomalyPath B in anomalyPath C in anomaly

700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900

(c) Anomalous traffic flow of paths

t

0

10

20

30

40

50

60

70

80

()

Percentage of path APercentage of path BPercentage of path C

700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

845

ndash900

900

ndash915

915

ndash930

(d) Percentage comparison

Figure 6 Effects of time intervals


Table 4 Hardwaresoftware configuration

Hardwaresoftware name VersionsizeServer 64-bitOperating system Windows Server 2008CPU 250GHzMemory 16Gb

Table 5 Average processing time for anomaly detection

Procedure name Time (s)GPS data transform (one day) 1917Wavelet transformPCA lt200Shewhart amp EWMA 232

respectively Each figure contains three subfigures withFigures 8(a) and 9(a) presenting the detection results of base-line 1 method Figures 8(b) and 9(b) presenting the detec-tion results of baseline 2 method and Figures 8(c) and 9(c)presenting the anomalous subregions detected using theproposed method

In the first case road reconstruction occurred on LiaoheRoad between 900 AM and 1100 AM on Jun 17 2012 Asshown in Figure 8 the red line presents the work zone and theorange region represents the detected anomalous subregionsIn Figures 8(a) and 8(b) the total areas of the anomaloussubregions around the work zone are small However usingthe detection results of the proposed method (as shown inFigure 8(c)) a larger collection of anomalous subregionswas obtained and all of the paths through these affectedsubregions can be determined In contrast with the resultsfrom the baseline methods our advisory paths can avoid theanomalous subregions that were not detected by the baselinemethods Thus the advisory paths can be more accurate anduseful for drivers or management departments to activelyavoid the anomalous subregions such as the black linesin Figure 8(c) These advisory paths can change the actualdriving routes of some vehicles and this effect can reduce thetraffic pressure in this area while accelerating the dissipationof anomalies

In the second case the proposed method detected atraffic anomaly near theHarbin International Conference andExhibition Center (HICEC) from 830 PM to 1000 PM onJul 30 2012 However this anomaly was not reported by thetraffic management department As shown in Figures 9(a)and 9(b) baseline 1 method cannot be used to detect anyanomalies around the HICEC (gray region) and baseline2 method can only detect a small region adjacent to theHICECHowever according to the daily news on the Internetthe Harbin International Automobile Industry Exhibition(HIAIE) was held in the HICEC The HIAIE is one of thelargest exhibitions in Harbin and can attract many dealerand automobile manufacturers that exhibit their productsThus a large number of citizens attend this grand exhibitionTo ensure safety the management department deploys manypolice officers in this area Thus the traffic anomalies inthis area may be ignored in the reports because it can be

0

2000

4000

6000

8000

10000

12000

14000

16000

Highway road Main road Minor road Slip road

Proc

essin

g tim

e (m

s)

Figure 7 Processing time for anomaly detection

assumed that this area is effectively controlledHowever goodcontrol does not mean that no traffic anomaly occurs Largetraffic pressure can result in short-term and large-scale trafficanomalies Thus the results of these two baseline methodsare not sufficient for supporting traffic management andemergency treatment However as shown in Figure 9(c) theproposed method detected a large-scale anomalous regionaround the HICEC which corresponds better with theactual traffic thus the accuracy of the proposed methodis much higher than the baseline methods Consequentlythe proposed method is more sensitive to short-term trafficanomalies and the development and dissemination of trafficanomalies can be controlled well by using the proposedmethod

4 Conclusions

A traffic anomalies detection method that uses taxi GPS datawas presented to explore one aspect of urban traffic dynamicsAnd a novel approach based on the distribution of traffic flowwas used for locating and describing traffic anomalies Thismethod provides an effective approach for discovering trafficanomalies between two adjacent regions The effectivenessand computing performance of this method were evaluatedby using a taxi GPS dataset of more than 15000 taxies forsix months in Harbin This method detected most of thereported anomalies because it combines the advantages of theShewhart control chart method and the EWMA control chartmethod Thus this method can detect the anomalies causedby rapidly changing traffic flows and slowly changing trafficflows According to the experimental results 7462 of theanomalies reported by the traffic administrative departmentwere identified which is much higher than the existingmethods based on LRT and PCA Compared with otheranomalies detectionmethods thismethod can identify trafficflows that cause traffic anomalies and provide effectivenessinformation for managers to solve traffic jam or emergencyresponse problems Furthermore this method can changethe granularity of region segmentation based on the actual


(a) Baseline 1 results (b) Baseline 2 results

(c) Proposed method results

Figure 8 Case 1 detection results




demand which satisfies the requirements of traffic anomaliesdetection for different purposes The average execution timeof this method is less than 10 seconds and the effectiveness ishigh enough to support real-time detection of anomalies


The authors declare no conflict of interests regarding thepublication of this paper

Acknowledgments

This research is supported by the National Natural ScienceFoundation of China (Project no 71203045) HeilongjiangNatural Science Foundation (Project no E201318) and theFundamental Research Funds for the Central Universities(Grant no HITKISTP201421) This work was performedat the Key Laboratory of Advanced Materials amp IntelligentControl Technology on Transportation Safety Ministry ofCommunications China


References

[1] B Pan Y Zheng D Wilkie and C Shahabi ldquoCrowd sensing oftraffic anomalies based on human mobility and social mediardquoin Proceedings of the 21st ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems(SIGSPATIAL rsquo13) pp 334ndash343 ACM New York NY USA2013

[2] Y Yue H-D Wang B Hu Q-Q Li Y-G Li and A G O YehldquoExploratory calibration of a spatial interaction model usingtaxi GPS trajectoriesrdquo Computers Environment and UrbanSystems vol 36 no 2 pp 140ndash153 2012

[3] Y Liu F Wang Y Xiao and S Gao ldquoUrban land uses andtraffic lsquosource-sink areasrsquo evidence from GPS-enabled taxi datain Shanghairdquo Landscape and Urban Planning vol 106 no 1 pp73ndash87 2012

[4] M Veloso S Phithakkitnukoon and C Bento ldquoUrbanmobilitystudy using taxi tracesrdquo in Proceedings of the InternationalWorkshop on Trajectory Data Mining and Analysis (TDMA rsquo11)pp 23ndash30 ACM September 2011

[5] C Chen D Zhang P S Castro et al ldquoReal-time detection ofanomalous taxi trajectories from GPS tracesrdquo in Mobile andUbiquitous Systems Computing Networking and Services pp63ndash74 Springer Berlin Germany 2012

[6] Y Ge H Xiong C Liu and Z-H Zhou ldquoA taxi driving frauddetection systemrdquo in Proceedings of the 11th IEEE InternationalConference on Data Mining (ICDM rsquo11) pp 181ndash190 December2011

[7] D Zhang N Li Z H Zhou et al ldquoiBAT detecting anomaloustaxi trajectories from GPS tracesrdquo in Proceedings of the 13thInternational Conference on Ubiquitous Computing pp 99ndash108ACM 2011

[8] J Zhang ldquoSmarter outlier detection and deeper understandingof large-scale taxi trip records a case study of NYCrdquo inProceedings of the ACM SIGKDD International Workshop onUrban Computing pp 157ndash162 ACM August 2012

[9] H Wang and R L Cheu ldquoA microscopic simulation modellingof vehicle monitoring using kinematic data based on GPS andITS technologiesrdquo Journal of Software vol 9 no 6 pp 1382ndash1388 2014

[10] J Yuan Y Zheng C Zhang et al ldquoT-drive driving directionsbased on taxi trajectoriesrdquo in Proceedings of the 18th SIGSPA-TIAL International Conference on Advances in Geographic Infor-mation Systems (GIS rsquo10) pp 99ndash108 ACM New York NYUSA November 2010

[11] B D Ziebart A L Maas A K Dey and J A BagnellldquoNavigate like a cabbie probabilistic reasoning from observedcontext-aware behaviorrdquo in Proceedings of the 10th InternationalConference on Ubiquitous Computing (UbiComp rsquo08) pp 322ndash331 ACM September 2008

[12] H Yoon Y Zheng X Xie and W Woo ldquoSmart itineraryrecommendation based on user-generated GPS trajectoriesrdquoin Ubiquitous Intelligence and Computing vol 6406 of LectureNotes in Computer Science pp 19ndash34 Springer Berlin Ger-many 2010

[13] J Yuan Y Zheng X Xie and G Sun ldquoDriving with knowledgefrom the physical worldrdquo in Proceedings of the 17th ACMSIGKDD International Conference on Knowledge Discovery andData Mining (KDD rsquo11) pp 316ndash324 ACM August 2011

[14] S Chawla Y Zheng and J Hu ldquoInferring the root cause in roadtraffic anomaliesrdquo in Proceedings of the 12th IEEE International

Conference on Data Mining (ICDM rsquo12) pp 141ndash150 December2012

[15] J A Barria and SThajchayapong ldquoDetection and classificationof traffic anomalies using microscopic traffic variablesrdquo IEEETransactions on Intelligent Transportation Systems vol 12 no3 pp 695ndash704 2011

[16] Q Chen Q Qiu H Li and Q Wu ldquoA neuromorphic archi-tecture for anomaly detection in autonomous large-area trafficmonitoringrdquo inProceedings of the 32nd IEEEACMInternationalConference on Computer-Aided Design (ICCAD rsquo13) pp 202ndash205 IEEE November 2013

[17] C Chen D Zhang P S Castro N Li L Sun and S Li ldquoReal-time detection of anomalous taxi trajectories from GPS tracesrdquoin Mobile and Ubiquitous Systems Computing Networkingand Services vol 104 of Lecture Notes of the Institute forComputer Sciences Social Informatics and TelecommunicationsEngineering pp 63ndash74 Springer Berlin Germany 2012

[18] Y Zheng Y Liu J Yuan and X Xie ldquoUrban computing withtaxicabsrdquo in Proceedings of the 13th International Conference onUbiquitous Computing pp 89ndash98 ACM September 2011

[19] W Liu Y Zheng S Chawla J Yuan and X Xie ldquoDiscoveringspatio-temporal causal interactions in traffic data streamsrdquo inProceedings of the 17th ACM SIGKDD International Conferenceon Knowledge Discovery and Data Mining (KDD rsquo11) pp 1010ndash1018 ACM New York NY USA August 2011

[20] Z Wang M Lu X Yuan J Zhang and H V D WeteringldquoVisual traffic jam analysis based on trajectory datardquo IEEETransactions on Visualization and Computer Graphics vol 19no 12 pp 2159ndash2168 2013

[21] T Sakaki M Okazaki and Y Matsuo ldquoEarthquake shakesTwitter users real-time event detection by social sensorsrdquo inProceedings of the 19th International Conference on World WideWeb (WWW rsquo10) pp 851ndash860 ACM April 2010

[22] E M Daly F Lecue and V Bicer ldquoWestland row why so slowFusing social media and linked data sources for understandingreal-time traffic conditionsrdquo in Proceedings of the 18th Interna-tional Conference on Intelligent User Interfaces (IUI rsquo13) pp 203ndash212 ACM March 2013

[23] V Chandola A Banerjee and V Kumar ldquoAnomaly detection asurveyrdquo ACM Computing Surveys vol 41 no 3 article 15 2009

[24] V J Hodge and J Austin ldquoA survey of outlier detectionmethodologiesrdquo Artificial Intelligence Review vol 22 no 2 pp85ndash126 2004

[25] L X Pang S Chawla W Liu and Y Zheng ldquoOn detection ofemerging anomalous traffic patterns using GPS datardquo Data ampKnowledge Engineering vol 87 pp 357ndash373 2013

[26] D Jiang P Zhang Z Xu C Yao and W Qin ldquoA wavelet-baseddetection approach to traffic anomaliesrdquo in Proceedings of the7th International Conference on Computational Intelligence andSecurity (CIS rsquo11) pp 993ndash997 December 2011

[27] A Gran and H Veiga ldquoWavelet-based detection of outliers infinancial time seriesrdquo Computational Statistics amp Data Analysisvol 54 no 11 pp 2580ndash2593 2010

[28] N J Yuan Y Zheng and X Xie ldquoSegmentation of urban areasusing road networksrdquo Tech Rep MSR-TR-2012-65 MicrosoftResearch 2012

[29] S G Mallat ldquoTheory for multiresolution signal decompositionthe wavelet representationrdquo IEEE Transactions on Pattern Anal-ysis and Machine Intelligence vol 11 no 7 pp 674ndash693 1989

[30] B R Bakshi ldquoMultiscale PCA with application to multivariatestatistical process monitoringrdquoAIChE Journal vol 44 no 7 pp1596ndash1610 1998


[31] A Lakhina M Crovella and C Diot ldquoDiagnosing network-wide traffic anomaliesrdquo ACM SIGCOMM Computer Communi-cation Review vol 34 no 4 pp 219ndash230 2004

[32] S Bersimis S Psarakis and J Panaretos ldquoMultivariate statisticalprocess control charts an overviewrdquo Quality and ReliabilityEngineering International vol 23 no 5 pp 517ndash543 2007

Research ArticleIdentifying Key Factors for Introducing GPS-Based FleetManagement Systems to the Logistics Industry

Yi-Chung Hu Yu-Jing Chiu Chung-Sheng Hsu and Yu-Ying Chang

Department of Business Administration Chung Yuan Christian University Chung Li District Taoyuan City 32023 Taiwan

Correspondence should be addressed to Yu-Jing Chiu yujingcycuedutw

Received 21 November 2014 Accepted 2 February 2015

Academic Editor Jinhu Lu

Copyright copy 2015 Yi-Chung Hu et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

The rise of e-commerce and globalization has changed consumption patterns Different industries have different logistical needsIn meeting needs with different schedules logistics play a key role Delivering a seamless service becomes a source of competitiveadvantage for the logistics industry Global positioning system-based fleet management system technology provides synergy totransport companies and achieves many management goals such as monitoring and tracking commodity distribution energysaving safety and quality A case company which is a subsidiary of a very famous food and retail conglomerate and operates thelargest shipping line in Taiwan has suffered from the nonsmooth introduction of GPS-based fleet management systems in recentyears Therefore this study aims to identify key factors for introducing related systems to the case company By using DEMATELand ANP we can find not only key factors but also causes and effects among key factors The results showed that support fromexecutives was the most important criterion but it has the worst performance among key factors It is found that adequate annualbudget planning enhancement of user intention and collaborationwith consultants with high specialty could be helpful to enhancethe faith of top executives for successfully introducing the systems to the case company

1 Introduction

The rise of e-commerce and globalization has changed con-sumption patterns Different industries have different logis-tical needs In meeting needs for small diverse and high-frequency pickups and deliveries at different locations indifferent packaging and according to different schedules andin determining how different operations such as purchasingmanufacturing warehousing distribution and managementcontribute to a good solution logistics play a key roleDelivering a seamless service has become a source of compet-itive advantage for the logistics industry Fleet managementsystems (FMS) have been available in the logistics industryfor many years Crainic and Laporte [1 2] pointed out thatfirst-generation FMS provided relatively simple functional-ities such as vehicle tracking components With increasedmanagement sophistication these systems have evolved intoplanning tools [3 4] In addition fleet management involvessupervising the use and maintenance of vehicles and asso-ciated administrative functions including coordination and

dissemination of tasks and related information to solve theheterogeneous scheduling and vehicle routing problem [5]For vehicle fleet management and monitoring one of themain applications is the global positioning system (GPS)technology [6 7] GPS-based fleet management system tech-nology has provided synergy to transport companies and hasachieved many management goals such as monitoring andtracking commodity distribution energy savings safety andquality A fleet management system is a complex network tomanage and control It is well known that most real-worldmanagement systems are typical complex and evolving net-works [8ndash11] and fleetmanagement systems are no exception

This research used the PTransport Company as an empir-icalstudy case The company which operates the largestshipping line in Taiwan is a subsidiary of a famous foodand retail conglomerate which is the largest group of chainstores in Taiwan The system had to serve the countryrsquoslargest logistics system and provide comprehensive logisticalsupport and fast supply to all outlets nationwide The PTransport Companywas committed to continuously enhance



the competitiveness by the introduction of GPS Althoughthe P Transport Companyworked energetically to implementintelligent fleet management systems these have not beensuccessful in recent years The P Transport Company wasin the system implementation phase at the time of thisresearch and wanted to avoid another failure in introducinga fleet management system After interviewing the managersof P Transport Company four main reasons for earlierfailures were identified organizational resistance to changeongoing information technology innovation lack of profes-sional training and experience in project staff and multiplecustomer patterns and complex operating procedures

This research intended to identify the key factors inintroducing GPS-based fleet management systems to thelogistics industry by the analysis of P Transport CompanyFor the purpose of this paper several factors were involvedand it was necessary to determine which of these factorswas the most significant for achieving the objective of thisstudy In addition this complex management problem wasa classic case of multiple-criteria decision-making (MCDM)and these indicators had interdependent impacts Regardingthe research methods analytic network process (ANP) is awidely usedmethod that considers interdependencies amongfactors and determines their relative importance [12ndash16]A combination of Decision-Making Trial and EvaluationLaboratory (DEMATEL) and ANP has been widely used tosolve various decision problems [17ndash20] To take interdepen-dencies into consideration and determine the key factors thispaper incorporates a novel combination of DEMATEL andANP into the study By analyzing the case company this studycontributes to explore an important issue that identifies keyfactors for introducing GPS-based fleet management systemsto the logistics industry using DEMATEL and ANP

The results showed that support from executives wasthe most important criterion and had profound influenceon other criteria Performance on other key factors wasimproved if corporate executives showed strong supportTheother key factors were user recognition funding and budgetproject team composition correct information in real timeand degree of completion of transmission equipment Theproposed model was implemented in a transport companyin Taiwan Based on the results obtained it was suggestedthat transport companies and the logistics industry introduceGPS-based fleet management systems which will increasetheir chance of success

Section 1 of this paper provides an introduction whichsummarizes the research motive purpose methodology andstudy results Section 2 provides a brief review of GPS-basedfleet management systems and key factors for introducingthese systems Section 3 describes the methodology usedand Section 4 presents an example and results Finallyconclusions and recommendations can be found in Section 5

2 Literature Review

21 Fleet Management Systems and GPS Intelligent trans-portation systems (ITS)were defined in [21] as using informa-tion technologies computers and communications in trans-portation systems to solve transportation problems These

systems increase transportation efficiency promote drivingsafety improve peoplersquos lives and raise industrial productivity[22] Fleet management systems (FMS) have been availablein the industrial domain such as the transport businessfor many years Currently these systems have evolved intocomplete enterprise management tools linking together allparts of the businessThe new trend clearly dictates increasedmanagement sophistication in terms of turning these toolsinto planning tools [3 4] They now include real-time assetmanagement focusing on current fleet locations and predic-tion of planned tasksThese systems today offer a broad rangeof functionalities including tight integration with internalenterprise resource planning (ERP) systems and systemslocated at customer sites Specifically extensive use of real-time data and wireless communications serve together withincreased intelligence for real-time planning where industrydevelopers identify these parameters as the primary driversfor current developments [23]

In an industrial context a complete logistics systeminvolves transporting rawmaterials from a number of suppli-ers delivering them to the factory for processing transport-ing the products to different depots and finally distributingthem to customers [5] In this case transportation for bothsupply and distribution requires effective management pro-cedures to optimize routes and costs These procedures formpart of the overall supply-chain management of the company[24] The American Heritage Dictionary defines a globalpositioning system as ldquoA system for determining a positionon the Earthrsquos surface by comparing radio signals fromseveral satellites Depending on your geographic location theGPS receiver samples data from up to six satellites it thencalculates the time taken for each satellite signal to reach theGPS receiver and from the difference in time of receptiondetermines your location [25]rdquo A number of literatureshave been published which provide information to engineersaboutGPS technology applications to transportation systemsespecially to intelligent transportation systems [26 27]

GPS became very important because not only did themilitary rely on them to provide navigation but the pub-lic sector did as well These devices were used by pilotsminers mountain climbers and many others working indangerous occupations [28] Several industries such as thelogistics realized this and started to focus on research andquality control These industries also realized the benefit ofcombining GPS technology with telecommunications Thisenabled GPS receivers to transmit data to a base stationfor analysis Another advance was a GPS architecture thatenabled integration of the technology into computers andother devices This opened up a huge spectrum of uses forGPS [28] Companies can reduce costs and create greatercustomer satisfaction by implementing GPS systems as partof already established processes [28] GPS became a ldquotool ofthe traderdquo in trucking companies for logistics management

GPS devices gave managers more accurate estimates ofboth the time of arrival and the time of delivery of goodsto the customer [29] As part of logistics managementfleet management can be a practical tool for managing avehicle fleet to improve scheduling operating efficiency andeffectiveness [30] In addition fleet management involves


Table 1 Aspects for the introduction of management information systems

Aspects Descriptions References

Organization

The impact of implementing a system in an organization the system must beaccepted by the organization and integrated into the workflow among other existinginformation systems Staff can have concerns arising from the nature of theorganizational change resistance mentality

[35ndash43]

Project base

The execution and management of the project IT project management must usuallywork with a series of complex problems and diverse staff In particular teammanagement requires a high degree of expertise to deal with project executionmanagement issues

[36 37 40 41 43]

Systemtechnology

Technical complexity of the system before building the system high-quality datamust be available The system must include information on whether the accuracytimeliness integration and flexibility of the technology can meet organizationalneeds

[35ndash43]

Consultants

Ability of enterprises to solve problems business consultants that have dealt with asimilar situation in the past can be expected to have specific experience andknowledge and to adapt solutions to the current problems encountered Thecapacity and performance of consultants during the project will affect the success orfailure of the entire project

[35ndash37 39]

Externalenvironment

Factors external to the organization for example the impact on the implementedsystem of external competitive pressures also refer to the impact of trade laws andregulations Industry competitive pressures and suppliers will affect allimplemented technologies

[38 42]

supervising the use and maintenance of vehicles and asso-ciated administrative functions including coordination anddissemination of tasks and related information to solveheterogeneous scheduling and vehicle routing problems [5]

22 Introduction of Management Information Systems Theintroduction of new systems can be understood from busi-ness experience and from the literature A successful systemintroduction provides positive benefits to an organizationbut a failed introduction can do harm to the organizationMany studies have focused on the key factors affectingthe introduction of a new system to a company Table 1summarizes related aspects and literatures for the intro-duction of management information systems and Table 2shows preliminary aspects and criteria cited from the relatedliteratures

3 Methodology

31 Delphi Method The Delphi method is a researchapproach to group decision-making Reference [31] indicatedthat the Delphi method depends on expertsrsquo experienceinstincts and values to determine outcomes In this methoda group of six experts discusses a specific question becauseexperts from different fields can be expected to providemultiple perspectives Besides the experts can understandeach otherrsquos perspectives in one round of the questionnaireand adjust their own perspectives in the next questionnaireround to reach consistency

The related operations are briefly introduced as followsFirst the appropriate experts are grouped according tothe nature of the question that must be decided Hence

the number of experts is determined in terms of the dimen-sions professional requirements complexity and scope ofthe problem In general the group will not exceed twentypeople Second background information about the decisionis transmitted to the experts and they are asked what elsethey need Furthermore they are advised of the questionsthat must be answered and any related requests Finallythe experts are asked to answer the questions in writingThird the experts indicate their perspectives and explain howthese perspectives were obtained from the information givenFourth the expert perspectives are synthesized for the firsttime to produce an information form which is sent to theexperts so that they can understand the differences betweentheir perspectives and those of others and adjust theirperspectives and evaluation accordingly Fifth themajor partof theDelphimethod involves collecting expertsrsquo perspectivesand providing feedback In other words the modified per-spectives from the experts are collected synthesized and sentback to each expert for further modification Note that eachexpertrsquos name is not included when the information is fedback to the experts as a group This process is repeated untilno expert submits further modifications Finally the expertsrsquoperspectives are synthesized and conclusions are presented

32 DEMATEL-Based ANP (DANP) Traditionally a net-work relation map (NRM) was necessary for ANP but NRMshould be acquired by other auxiliary tools UndoubtedlyDecision-Making Trial and Evaluation Laboratory (DEMA-TEL) is an appropriate choice for constructing NRM [20]by describing interdependencies visually in the form ofnetworks consisting of explainable nodes and directed arcs[31] Nevertheless a serious problem for ANP is that ifthere are too many criteria involving pairwise comparisons


Table 2 Preliminary aspects and criteria for the study

Aspects Criteria Descriptions

Organization

Top executives supportExecutivesrsquo subjective preferences or understanding of the project continuedparticipation promises of funding and resources required and removal ofobstacles to the project

Enterprise process reengineering The need to change the organizationrsquos structure responsibilities and workflowin response to the implemented system

User recognition Whether employees have sufficient momentum to drive their participation inthe system

Funding and budget The project budget for implementing software hardware and subsequentmaintenance requirements

Project base

Clear objectives A clear understanding of importing goals and performance those are from thevarious departments

Project team composition Organizations with outstanding staff from ministries can take up thechallenge and work together to resolve difficulties

Project management andmonitoring Project leaders and teams control project progress

Effective communication To resolve conflictEducation and training Actual effectiveness of education and training

Systemtechnology

Timely and correct information Control over correct and timely input informationDegree of difficulty in softwareand hardware maintenance

Degree of maintenance difficulty for system and hardware devices in thefuture

Degree of difficulty in technologysetup

Degree of difficulty in setup of system technology and extension to variouscenters

Degree of completeness oftransmission equipment Transmission performance and scalability of equipment installed in a truck

Consultant

Experience of consultants Industrial familiarity expressive ability and communication skills ofconsultants

Ability of consultants Degree of professional competence of consultants for each module in thesystem

Coordination andcommunication

Service gap between expectation and perception of customers in theconsultantrsquos interaction process

Externalenvironment

Industry competitive pressureDevelopment of innovation in industry is very rapid and therefore whenfacing competition a further assessment of the competitive environmentfacing the enterprise is required

Customer acceptance Willingness of customers to implement a system and conditions imposed

then the time required for pairwise comparisons increasessubstantially Moreover it is not easy to achieve consistency[32] especially for the matrix with high order because ofthe influence of the limited ability of human thinking and theshortcomings of one to nine scale [33] To solve the above-mentioned problems the so-called DANP took the totalinfluence matrix generated by DEMATEL as the unweightedsupermatrix of ANP directly to avoid troublesome pairwisecomparisons Similar to ANP relative weights of individualfactors can be obtained by generating a limiting supermatrixTzeng and Huang [20] introduced the complete frameworkof DANP

In particular the framework of DANP used in this paperhas several distinct features compared to [20] First this paperconsiders prominences generated by DEMATEL and relativeweights generated by DANP at the same time to determinekey factors instead of using relative importance by DANPmerely In other words as represented by dashed lines in

Figure 1 both DEMATEL and DANP have the power tovote for key factors Second we focus on the causal diagramfor key factors rather than all factors Moreover an arc isdirected from one factor to another one if the former has thegreatest influence on the latter This can simplify greatly therepresentation of a causal diagram and facilitate the analysisof interdependence among key factors Besides the causaldiagram is not dependent on relation of each factor Thereason is that the greater the relation of a factor is the greaterthe influence of it on another factor is not assured Such anovel variant of the traditional DANP is briefly depicted inFigure 1

321 Determining the Total Influence Matrix The perfor-mance values used to represent the degree of influence ofone element on another were 0 (no effect) 1 (little effect) 2(some effect) 3 (strong effect) and 4 (certain effect) Next thedirect influence matrix Z was constructed using the degree


Acquire a direct influence matrix (Z)

Normalized Z(X)

Generate a total influence matrix (T)

Determinerelation of each factor

Determine prominence of

each factor

Depict a causal diagram for all factors

Determine key factors

Depict a causal diagram for key factors Form an unweighted supermatrix

Construct a weighted supermatrix

Generate a limiting supermatrix

Find relative weights

DEMATEL

ANP

Figure 1 The proposed framework of DANP

of effect between each pair of elements as obtained by thequestionnaire 119911

119894119895represents the extent to which criterion 119894

affects criterion 119895 All diagonal elements are set to zero

Z =

[[[[[[[

[

1199111111991112sdot sdot sdot 119911

1119899

1199112111991122sdot sdot sdot 119911

2119899

11991111989911199111198992sdot sdot sdot 119911

119899119899

]]]]]]]

]

(1)

Thedirect influencematrixZwas subsequently normalized toyield a normalized direct influence matrixX after calculating

120582 =

1

max1le119894le119899sum119899

119895=1119885119894119895

(119894 119895 = 1 2 119899)

X = 120582 sdot Z(2)

The formula (T = X(I minus X)minus1) was used to represent thetotal influencematrixT after normalizing the direct influencematrix In this step O was the zero matrix and I the identitymatrix

lim119870rarrinfin

X119870 = 0

119879 = lim119909rarrinfin(X + X2 + sdot sdot sdot + K119896) = X (IminusX)minus1

(3)

The total influence matrix T was viewed as an unweightedsupermatrix and was used to normalize the total influencematrix to obtain the weighted matrix W for ANP FinallyW was multiplied by itself several times until convergence to

obtain the limiting supermatrixWlowast and the global weight ofall elements Below a simple example is used to illustrate theabovementioned operations with respect to factors 119860 119861 119862and119863 for a decision problem Let a direct influence matrix Zbe obtained as follows

Z =119860

119861

119862

119863

((

(

119860

0

3

3

3

119861

2

0

1

2

119862

2

2

0

2

119863

2

1

2

0

))

)

(4)

This matrix was subsequently normalized to obtain thenormalized relationmatrixXThen the total influencematrixT was calculated using X(I minus X)minus1

X =119860

119861

119862

119863

((

(

119860

0000

0337

0326

0337

119861

0233

0000

0116

0198

119862

0279

0198

0000

0198

119863

0233

0116

0244

0000

))

)

T =

119860

119861

119862

119863

(

119860

0628

0817

0839

0876

119861

0580

0356

0483

0559

119862

0691

0593

0449

0637

119863

0615

0493

0605

0424

)

119889

2513

2259

2377

2497

119903 3159 1979 2370 2137

(5)

Each row of the total influence matrix was summed toobtain the value of 119889 and each column of the total influencematrix was summed to obtain the value of 119903 Hence the sumof every row plus the sum of every column (ie 119889 + 119903) calledthe prominence shows the relational intensity of the elementin questionThe greater the prominence becomes the greaterthe degree of importance will be among factors The sum ofevery rowminus the sum of every column (119889minus119903) is called therelation If the relation is positive then the element is inclinedto affect other elements actively andwas referred to as a causeIf the relation is negative the element is inclined to be affectedby other elements and was referred to as an effect In otherwords a positive relation means the degree to which such afactor affected the others is inclined to be stronger than thedegree to which it was affected [17] (see Table 3)

The total influence matrix was then normalized to obtainthe weighted supermatrixW (see Table 4)

Finally W was multiplied by itself several times untilconvergence to obtain the limiting supermatrix Wlowast Factors119861 119862 and 119863 can be categorized into a class of ldquocauserdquo Itis worthy to mention that although the relation of factor119863 is the most positive (ie 03598) it has not the greatestinfluences on factors 119860 119861 and 119862 For instance factor 119860which can be categorized into a class of ldquoeffectrdquo imposes thegreatest influence on factor 119862 (ie 0691) rather than 119863 (ie0637)


Table 3

Factor 119889 119903 119889 + 119903 Ranking 119889 minus 119903

119860 2513 3159 5673 1 minus06462119861 2259 1979 4238 4 02796119862 2377 2370 4746 2 00068119863 2496 2137 4633 3 03598

Table 4

119860 119861 119862 119863

119860 0199 0293 0291 0288119861 0259 0180 0250 0231119862 0266 0244 0190 0283119863 0277 0283 0269 0199

322 Identifying Key Factors Following the simple examplein the previous subsection the comparative weights of ele-ments 119860 119861 119862 and119863 were determined as 0266 0231 0246and 0256 respectively However it can be seen that the rank-ings of the importance for factors resulting fromprominencesgenerated by DEMATEL and relative weights obtained byDANP were inconsistent In our opinion since both DEMA-TEL and DANP provide partial messages regarding theselection of key factors decisions on key factors shouldnot be based on prominences generated by DEMATEL orrelative weights obtained by DANP as the sole considerationThis motivates us to use the abovementioned message todetermine the final importance rankings of factors Theoverall rankings for factors are shown in Table 5 by arrangingthe sum of rankings of each factor in ascending order

323 Depicting the Causal Diagram for Key Factors Follow-ing the previous subsection we can depict a causal diagramfor key factors For example because factors119860119862 and119863werekey factors the total influence matrix was used to draw acausal diagram The total influence matrix showed that thefactors affecting 119860 119862 and 119863 most strongly were still 119860 119862and119863 (see Figure 2)

Then a causal diagram with respect to factors 119860 119862 and119863 can be easily depicted as shown in Figure 3

As shown in the causal diagram interactions existedbetween factors 119860 119862 and 119863 Moreover it is reasonablefor managers to get down to performance improvement of119860 or 119863 for the problem energetically For 119860 performanceimprovement of 119860 can facilitate those of 119862 and 119863 Howeversince 119860 is categorized into a class of ldquoeffectrdquo the performanceof 119863 is usually undertaken to improve at first to promotethe performance improvement of the other key factors Wethink that whether 119860 can be taken as a starting point or notshould be dependent on the real situation That is ldquocauserdquoor ldquoeffectrdquo is just for reference The importance-performanceanalysis (IPA) formulated by Martilla and James [34] can bean appropriate tool to help users examine key factors that arenecessary to be improved

Table 5

Factors DEMATEL DANP Sum ofrankings

Overallrankings

119860 1 1 2 1119861 4 4 8 4119862 2 3 5 2119863 3 2 5 2We can take factors 119860 119862 and119863 as key factors

A B C DA 0628 0580 0691 0615B 0817 0256 0593 0493C 0839 0483 0449 0605D 0876 0559 0637 0424

T =

Figure 2

DA

C

Figure 3

4 Empirical Study

41 Case Introduction P Transport Company a companyowned by a large corporation operates the largest freighttransportation line in Taiwan Their fleet consists of 1700trucks and is capable of serving more than 5000 retailstores The company was beginning to introduce electronicoperations and systems to enhance its competitiveness inthe industry and to achieve the goals given by the cor-poration in the hope that these systems would lead tohigher corporate operating efficiency However the resultswere often unsatisfactory P Transport Companyrsquos recentattempt to introduce an intelligent fleet management systemwas not successful Their testing and startup costs exceededNT 10 million with more than several dozen test vendorsAfter discussion with company managers the reasons forthe earlier implementation failure were identified as followsaccumulated organizational cost considerations resistancefrom employees to innovative changes lack of professionalknow-how and experience in the project team ongoinginformation technology innovation and evolution and mul-tiple patterns of customers and job complexity leading todifficulties in system development

42 Determining the Formal Decision Structure Most of thedecision-makers made their system implementation deci-sions based on their subjective views and various working


Table 6 A formal decision structure for the case study


Organization(119860)

Top executives support (1198601)Executivesrsquo subjective preferences or understanding of the project continuedparticipation promises of funding and resources required and removal ofobstacles to the project

User recognition (1198602) Whether employees have sufficient momentum to drive their participation inthe system

Funding and budget (1198603) The project budget for implementing software hardware and subsequentmaintenance requirements

Project base (119861)

Project team composition (1198611) Organizations with outstanding staff from ministries can take up thechallenge and work together to resolve difficulties

Project management andmonitoring (1198612) Project leaders and teams control project progress

Education and training (1198613) Actual effectiveness of education and training

Systemtechnology (119862)

Timely and correct information(1198621) Control over correct and timely input information

Degree of difficulty in softwareand hardware maintenance (1198622)

The degree of maintenance difficulty for the system and for hardware devicesin the future

Degree of completeness oftransmission equipment (1198623) Transmission performance and scalability of equipment installed in a truck

Externalenvironment(119863)

Experience and ability ofconsultants (1198631)

Industrial familiarity expressive capability and communication skills of theconsultant Level of professional competence of the consultant for eachmodule in the system

Coordination andcommunication (1198632)

Because the development of industry innovation is very rapid when facingcompetition a further assessment of the competitive environment facing theenterprise is required

Customer acceptance (1198633) Willingness of customers to implement a system and conditions imposed

rules This approach was likely to lead to wrong decisionsTo determine how to reduce the risk of failure an objectiveand quantitative approach was required to help companiesidentify the key factors in successful system introductionThe P Transport Company was selected for this researchas an empirical case to illustrate how to identify the keyfactors in introducing aGPS-based fleetmanagement systemA survey was carried out to collect expertsrsquo perceptionsinvolving six managers from the P Transport Company whowere involved in logistics and who had system softwaredevelopment experience

35 aspects and 144 criteria were identified after a literaturereview All these indicators were integrated according to sim-ilarities in definition and semantics and five aspects and 18criteria were selected for the prototype research architectureTo increase the possibility of success in implementing theGPS-based fleet management system the Delphi methodwas used in this study to revise the prototype architectureinto a formal decision structure as shown in Table 6 It wasfound that the consensus deviation index (CDI) in the Delphimethod of each factor is lower than 01 after the third roundand four aspects and 12 criteria were thus considered in thefinal evaluation framework Note that CDI is used to indicatethe degree of the common consensus of consults The greaterthe CDI is the worse the common consensus will be Thequestionnaire required by DEMATEL was designed and tenqualified managers from the P Transport Company wereinvited to provide their opinions

43 Result Analysis

431 Importance Analysis for Aspects Based on the expertsurvey and the DEMATEL method the initial direct influ-ence matrix for aspects was calculated using (1) with theresults shown in Table 7 The normalized direct influencematrix was obtained using (2) with the results shown inTable 8 The total influence matrix was calculated using (3)with the results shown in Table 9 The prominence andrelation of each aspect are shown in Table 10

As shown in Table 11 a weighted supermatrix can beobtained by normalizing the total influence matrix Thelimiting supermatrix derived by the weighted supermatrixwas shown in Table 12

The overall rankings for aspects are shown in Table 13 byarranging the sum of rankings of each aspect in ascendingorder It is clear that ldquoOrganizationsrdquo is the most importantaspect According to the total influence matrix for aspects acausal diagram depicted in Figure 4 shows that P TransportCompany should energetically get down to performanceimprovement of ldquoOrganizationsrdquo to facilitate those of theother aspects Also it is reasonable for P Transport Companyto undertake the development of appropriate strategies forimproving ldquoOrganizationsrdquo because ldquoOrganizationsrdquo is cate-gorized into a class of ldquocauserdquo It is noted that the proposedcausal diagram does not make use of prominences andrelations This is quite different from the traditional causaldiagram


Table 7 The initial direct influence matrix for aspects

Aspects 119860 119861 119862 119863

119860 00000 20000 24000 20000119861 29000 00000 17000 10000119862 28000 10000 00000 21000119863 29000 17000 17000 00000

Table 8 The normalized direct influence matrix for aspects

Aspects 119860 119861 119862 119863

119860 00000 02326 02791 02326119861 03372 00000 01977 01163119862 03256 01163 00000 02442119863 03372 01977 01977 00000

Table 9 The total influence matrix for aspects

Aspects 119860 119861 119862 119863 119889

119860 06278 05803 06905 06146 25132119861 08166 03563 05933 04925 22587119862 08389 04832 04492 06052 23765119863 08761 05593 06366 04242 24963119903 31593 19791 23697 21365

Table 10 Prominence and relation of each aspect

Aspects 119889 119903 119889 + 119903 119889 minus 119903

119860 25132 31593 56725 minus06462119861 22587 19791 42378 02796119862 23765 23697 47461 00068119863 24963 21365 46328 03598

Table 11 The weighted supermatrix for aspects

Aspects 119860 119861 119862 119863

119860 01987 02932 02914 02877119861 02585 01800 02504 02305119862 02655 02442 01896 02832119863 02773 02826 02686 01986

Table 12 The limited supermatrix for aspects

Aspects 119860 119861 119862 119863

119860 02662 02662 02662 02662119861 02312 02312 02312 02312119862 02464 02464 02464 02464119863 02562 02562 02562 02562

432 Importance Analysis for Criteria Based on the expertsurvey and the use of the DEMATEL method the initialdirect influence matrix in Table 14 for criteria was calculatedusing (1) The normalized direct influence matrix in Table 15was obtained through (2) The total influence matrix inTable 16 was calculated using (3) Table 17 summarizesthe prominence and relation of each criterion Table 18

Table 13 The overall ranking for aspects

Aspects DEMATEL DANP Sum ofrankings

Overallrankings

Organizations (119860) 1 1 2 1Project base (119861) 4 4 8 3System technology(119862) 2 3 5 2

Externalenvironment (119863) 3 2 5 2

Organizations(A)

External environment

(D)System

technology (C)

Project base (B)

Figure 4 The causal diagram for aspects

summarizes the causeeffect properties of twelve criteriaconsidered


The overall rankings for criteria are shown in Table 21 byarranging the sum of rankings of each criterion in ascend-ing order According the overall ranking list we take topexecutive support (1198601) funding and budget (1198603) experienceand ability of consultant (1198631) project team composition (1198611)timely and correct information (1198621) degree of completenessof transmission equipment (1198623) and user recognition (1198602)as key criteria

433 Importance-Performance Analysis To assess the cri-terion performances ten managers (1198781 1198782 11987810) fromthe P Transport Company were invited as survey subjectsThe relationship between rating and performance shown inTable 22 was also provided to subjects The average values forthe ten managers regarding performance on twelve criteriaare shown in Table 23 After consulting ten experts they allagreed to use 75 as a threshold value to distinguish criteriawith acceptable (ge75) or unacceptable (lt75) performancevalues from twelve criteria Each criterion with its rank andperformance value is depicted in Figure 5 which is used byIPA to examine which key factors should be concentrated

From Figure 5 it can be seen that in addition to topexecutive support (1198601) and funding and budget (1198603) fivekey criteria such as timely and correct information (1198621) anddegree of completeness of transmission equipment (1198623) fallinto the upper right grid P Transport Company should keepup the good performances of those key factors that fall intosuch a grid Also P Transport Company must effectivelyimprove the performances of top executive support and


Table 14 The initial direct influence matrix for criteria

Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 40000 40000 40000 24000 20000 28000 40000 20000 40000 30000 400001198602 30000 00000 20000 18000 22000 20000 30000 00000 00000 00000 30000 200001198603 39000 20000 00000 30000 19000 21000 24000 25000 25000 36000 20000 220001198611 16000 27000 30000 00000 19000 30000 23000 20000 10000 17000 40000 290001198612 10000 16000 10000 10000 00000 30000 24000 10000 20000 24000 26000 180001198613 01000 15000 12000 02000 00000 00000 21000 00000 01000 04000 10000 140001198621 20000 18000 20000 14000 16000 10000 00000 30000 00000 00000 10000 300001198622 10000 10000 25000 14000 18000 19000 27000 00000 20000 25000 15000 140001198623 25000 20000 29000 20000 19000 20000 26000 30000 00000 29000 10000 200001198631 30000 30000 30000 08000 23000 30000 24000 00000 00000 00000 40000 300001198632 29000 20000 00000 06000 16000 26000 21000 09000 00000 31000 00000 130001198633 18000 13000 14000 02000 09000 03000 10000 00000 00000 00000 18000 00000

Table 15 The normalized direct influence matrix for criteria

Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 01105 01105 01105 00663 00552 00773 01105 00552 01105 00829 011051198602 00829 00000 00552 00497 00608 00552 00829 00000 00000 00000 00829 005521198603 01077 00552 00000 00829 00525 00580 00663 00691 00691 00994 00552 006081198611 00442 00746 00829 00000 00525 00829 00635 00552 00276 00470 01105 008011198612 00276 00442 00276 00276 00000 00829 00663 00276 00552 00663 00718 004971198613 00028 00414 00331 00055 00000 00000 00580 00000 00028 00110 00276 003871198621 00552 00497 00552 00387 00442 00276 00000 00829 00000 00000 00276 008291198622 00276 00276 00691 00387 00497 00525 00746 00000 00552 00691 00414 003871198623 00691 00552 00801 00552 00525 00552 00718 00829 00000 00801 00276 005521198631 00829 00829 00829 00221 00635 00829 00663 00000 00000 00000 01105 008291198632 00801 00552 00000 00166 00442 00718 00580 00249 00000 00856 00000 003591198633 00497 00359 00387 00055 00249 00083 00276 00000 00000 00000 00497 00000

Table 16 The total influence matrix for criteria

Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633 119889

1198601 01250 02233 02211 01894 01618 01718 02066 01854 01023 02070 02120 02347 224041198602 01424 00664 01129 00954 01090 01150 01484 00500 00274 00582 01475 01249 119751198603 01991 01544 01007 01508 01311 01526 01722 01371 01064 01808 01621 01682 181551198611 01294 01542 01563 00593 01173 01606 01537 01094 00602 01181 01938 01663 157861198612 00915 01064 00878 00699 00504 01407 01334 00697 00753 01158 01356 01170 119361198613 00316 00647 00553 00240 00212 00230 00828 00183 00112 00296 00533 00655 048041198621 01085 01029 01082 00795 00883 00807 00629 01188 00273 00512 00885 01398 105671198622 00962 00947 01311 00855 01019 01164 01447 00487 00806 01242 01120 01116 124771198623 01521 01393 01621 01165 01205 01368 01635 01403 00376 01511 01215 01482 158951198631 01614 01602 01518 00802 01243 01561 01513 00561 00320 00695 01910 01665 150021198632 01319 01132 00593 00575 00890 01249 01196 00625 00217 01277 00654 01007 107341198633 00816 00679 00671 00315 00508 00399 00624 00252 00143 00309 00824 00359 05899119903 14507 14476 14136 10395 11656 14185 16015 10217 05964 12641 15651 15790

funding and budget that fall into the upper left grid Ofcourse1198601 and1198603 would pose a serious threat to P TransportCompany if they are ignored Also resources committedto those criteria that fall into lower right grid would bebetter employed elsewhere and it is not necessary to focusadditional effort on 1198622

According to the total influence matrix in Table 13 acausal diagram depicted in Figure 4 shows that P TransportCompany should energetically get down to performanceimprovements of top executive support (1198601) and funding andbudget (1198603) for introducing GPS-based fleet managementsystems to facilitate those of the other key factors Also


3

Impo

rtan

ce ra

nkin

g

Noncritical

Critical1

7

8

12

50 55 60 65 70 75 85 9580 90 100Performance value

Concentrate here Key up the good work

Possible overkillLow priority

Experience and ability of consultants (D1)

Project team composition (B1)

Timely and correct information (C1)

Degree of difficulty in software and hardware maintenance (C2)

Customer acceptance (D3)

Project management and monitoring (B2)

Coordination and communication (D2)

Education and training (B3)

Top executives support (A1)

Funding and budget (A3)

User recognition (A2)

Complete degree of transmission equipment (C3)

Figure 5 IPA for evaluation criteria

Table 17 Prominence and relation of each criterion

Criteria 119889 119903 119889 + 119903 119889 minus 119903

1198601 22404 14507 36911 078971198602 11975 14476 26451 minus025001198603 18155 14136 32291 040181198611 15786 10395 26181 053901198612 11936 11656 23592 002801198613 04804 14185 18990 minus093811198621 10567 16015 26582 minus054481198622 12477 10217 22694 022601198623 15895 05964 21860 099311198631 15002 12641 27643 023621198632 10734 15651 26386 minus049171198633 05899 15790 21689 minus09891

the selection of 1198601 and 1198603 to be the start is very appropriatebecause they are categorized into a class of ldquocauserdquo Toimprove 1198601 effectively executives of P Transport Companyshould promise that they must continue participation pro-vide funding and resources required and remove obstaclesactively to the project for the introduction of GPS-based fleetmanagement systems As for performance improvement of1198603 P Transport Company should provide adequate budgetfor implementing the software hardware and subsequentmaintenance requirements In Figure 6 it can be seen that1198601 and 1198603 influenced each other This means that adequateannual funding and budget planning are necessary in thelong term so as to enhance the faith of top executivesfor successfully introducing the information systems to PTransport Company As in the previous subsection theproposed causal diagram is a kind ofNRManddoes notmakeuse of prominences and relations

Since the improvement of 1198601 with the worst rating isurgent for P Transport Company in addition to 1198603 itis interesting to explore whether other factors can havecertain influence on 1198601 The total influence matrix showsthat 1198603 has the greatest impact on 1198601 and key criteria1198631 1198623 and 1198602 have the second the third and the forthgreatest impacts respectively It is reasonable to speculate thatenhancement of intention of using the systems for employeesand collaboration with consultants with high specialty can behelpful to enhance the support of executives In Figure 6 theformer and the latter impacts on 1198601 coming from 1198602 and1198631are indicated as dashed lines The abovementioned strategiesfor 1198601 and 1198603 can concretely implement the improvementof ldquoOrganizationsrdquo It is suggested that leverage of the totalinfluence matrix and the causal diagram could help usdevelop strategies of improvement in key factors especiallyfor those falling into the upper left grid in IPA Such ananalysis has its potentiality of being widely applied to otherproblem domains

5 Conclusions

Intelligent transportation systems have been in operationfor many years and commercial vehicle operation issueshave become important ITS trends in many developedcountries GPS-based fleet management systems are veryimportant to the logistics industry especially in transportcompaniesThese systems canmonitor and track commoditydistribution thus saving energy Moreover they also improvescheduling operating efficiency and effectiveness Becausefleet management systems are very important the successfulintroduction of these systems has become a key issue

The purpose of this research was to identify the keyfactors for introducing GPS-based fleet management systemsto transport companies DEMATEL andANPwere combined


Table 18 Causeeffect properties of criteria

Causeeffect Criteria

CauseTop executives support (1198601) funding and budget (1198603) project team composition (1198611) project management andmonitoring (1198612) degree of difficulty in software and hardware maintenance (1198622) complete degree of transmissionequipment (1198623) and experience and ability of consultants (1198631)

Effect User recognition (1198602) education and training (1198613) timely and correct information (1198621) coordination andcommunication (1198632) and customer acceptance (1198633)

Table 19 The weighted supermatrix for criteria

1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00862 01542 01564 01822 01388 01211 01290 01815 01715 01637 01355 014861198602 00982 00459 00799 00917 00935 00810 00927 00490 00459 00461 00943 007911198603 01372 01066 00712 01451 01125 01076 01075 01342 01784 01430 01036 010651198611 00892 01065 01105 00570 01007 01132 00960 01071 01009 00934 01238 010531198612 00631 00735 00621 00673 00432 00992 00833 00682 01263 00916 00866 007411198613 00218 00447 00391 00230 00182 00162 00517 00179 00188 00234 00341 004151198621 00748 00711 00765 00765 00757 00569 00393 01163 00458 00405 00566 008851198622 00663 00654 00927 00822 00874 00821 00904 00477 01352 00983 00716 007071198623 01048 00963 01147 01121 01034 00965 01021 01374 00630 01195 00776 009381198631 01112 01106 01074 00771 01066 01101 00945 00549 00537 00549 01220 010541198632 00909 00782 00420 00554 00764 00880 00747 00612 00364 01011 00418 006381198633 00562 00469 00474 00303 00436 00281 00390 00247 00240 00245 00527 00227

Table 20 The limited supermatrix for criteria

1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 014691198602 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 007491198603 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 012381198611 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 009801198612 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 007661198613 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 002851198621 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 006871198622 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 008381198623 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 010311198631 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 009061198632 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 006661198633 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386

Table 21 The overall ranking for criteria

Criteria DEMATEL DANP Sum of rankings Overall rankingsTop executives support (1198601) 1 1 2 1User recognition (1198602) 5 8 13 5Funding and budget (1198603) 2 2 4 2Project team composition (1198611) 7 4 11 4Project management and monitoring (1198612) 8 7 15 8Education and training (1198613) 12 12 24 12Timely and correct information (1198621) 4 9 13 5Degree of difficulty in software and hardware maintenance (1198622) 9 6 15 8Degree of completeness of transmission equipment (1198623) 10 3 13 5Experience and ability of consultants (1198631) 3 5 8 3Coordination and communication (1198632) 6 10 16 10Customer acceptance (1198633) 11 11 22 11


Table 22 Relationship between rating and performance

Rating 0 25 50 75 100Performance Very dissatisfied Dissatisfied Ordinary Satisfied Very satisfied

Table 23 Performance assessment of twelve criteria

Criteria Subjects Average1198781 1198782 1198783 1198784 1198785 1198786 1198787 1198788 1198789 11987810

Top executives support (1198601) 60 65 65 65 60 60 55 65 65 50 61User recognition (1198602) 85 80 70 75 75 65 80 75 80 70 76Funding and budget (1198603) 75 75 60 75 80 75 60 60 65 70 70Project team composition (1198611) 90 95 85 85 90 90 90 85 95 95 90Project management and monitoring (1198612) 80 75 80 75 85 75 80 90 90 80 81Education and training (1198613) 80 80 80 90 85 75 80 80 90 90 83Timely and correct information (1198621) 85 80 90 90 85 90 80 85 80 80 85Degree of difficulty software andhardware maintenance (1198622) 70 75 65 75 80 75 60 60 70 70 70

Complete degree of transmissionequipment (1198623) 90 95 85 90 90 90 90 85 95 85 90

Experience and ability of consultant (1198631) 75 75 75 80 80 80 75 70 70 75 76Coordination and communication (1198632) 70 75 80 85 80 75 70 80 80 70 77Customer acceptance (1198633) 80 75 70 75 75 70 80 75 80 70 75

to determine the key indicators identify the most importantone and discover how it affects others Top executive supportwas determined to be the most important criterion in thisstudy other key factors selected were funding and budgetexperience and ability of consultants project team composi-tion user recognition timely and correct information anddegree of completeness of transmission equipment Theseseven key factors are discussed below

Large organizations cannot avoid bureaucratic culturesand egos The introduction of new technologies and systemswill replace existing modes of operation often leading toresistance from conservative older employees and execu-tives who are unwilling to change The functioning of theorganization from the financial technical and training unitsto the business units determines the success or failure ofa system introduction Only executives can formulate top-down requirements and determine that system implementa-tion becomes a clear policy objective before they can driveinnovation across the enterprise

In the case of enterprises with limited resources imple-menting a new system requires large amounts of fund-ing time and human resources which are not necessarilyproportional to the rate of return that can be obtainedThis reality makes executives and shareholders conservativeBefore implementing a system a large budget must be setaside which will affect the current year net income and afterimplementation system maintenance costs will continue aslong-term operating costs Implementing new systems isclosely related to funding and only executives can set asidebudgets whereas the company has the resources for systemdevelopment and implementation

Implementing new technology and systems is not originalbusiness expertise and relies heavily on the technologyand experience of manufacturers to avoid costly mistakesLarge organizations are looking for manufacturers with well-oiled operations and similar size to ensure system operationand maintenance Therefore the experience and ability ofconsultants are important to enterprises The composition ofthe project team has a major impact on successful systemimplementation Members must have expertise in varioussectors to fully express the operating system requirementsof different departments thus facilitating interagency com-munication and coordination and helping system specifi-cation and development Innovation is not only driven byexecutives but requires the cooperation of all All usersmust accept change modify habits and adopt new operatingprocedures to enhance operational effectiveness A new GPSsystem has been developed which aims to achieve mapdatabase integration including real-time control data relatedto vehicle dynamics and driving speed braking emergencydeceleration arrival time temperature recording and otherimportant management information Timely and correctsystem output is the basic requirement for the transportcompany

The transmission equipment implemented for this GPSsystem features a link through the carrsquos transmission totransmit relevant information back to the company Based onthe current distinction between 2G and 3G a 3G system withintegrated touch screen and built-in CPU and memory waschosen for this project It was able to collect data on a deviceand send it through the devicersquos built-in program modulewithout preprocessingThe informationwas then transmitted




Key factorsUser recognition (A2) Funding and budget (A3)








Degree of difficulty in software and hardware

maintenance (C2)

Figure 6 The causal diagram for evaluation criteria

over a 3G link to the background avoiding too heavy burdenon this background to enhance the availability of accuratereal-time information

For the transport industry traffic accidents are the maincauses of violations caused by domestic carriers Manycasualties of trucks occurred in the past and have tended toplace less emphasis on the implementation of GPS-based fleetmanagement systems Actually violations can be reducedwith successful implementation of a system to avoid socialharm Abnormal driving behavior will become apparentthrough the fleet management system (speed travel timedriving illegal routes etc) and a temperature control featurewill be available in real time to prevent excessive heatingor cooling during delivery of goods ensuring food safetyThese research results can be used by the logistics industryto implement a GPS-based fleet management system As forfactory management logistics operators can also be used asan important reference for future systems before importingdataThe systemwill also provide opportunities to learn fromothers in the transport sector thereby enhancing the overallquality of transportation services



Acknowledgments

The authors would like to thank the anonymous referees fortheir valuable commentsThis research is partially supportedby the National Science Council of Taiwan under Grant noNSC 102-2410-H-033-039-MY2

References

[1] T G Crainic and G Laporte Fleet Management and LogisticsKluwer Academic Publishers Boston Mass USA 1998

[2] J Mele ldquoFleet management systems the future is hererdquo FleetOwner vol 100 no 8 p 88 2005

[3] T McLoad Fleet Management SystemsThe Future is Here FleetOwner 2005

[4] R van der Heijden and V Marchau ldquoInnovating road trafficmanagement by ITS a future perspectiverdquo International Journalof Technology Policy and Management vol 2 no 1 pp 20ndash392002

[5] C G Soslashrensen and D D Bochtis ldquoConceptual model of fleetmanagement in agriculturerdquo Biosystems Engineering vol 105no 1 pp 41ndash50 2010

[6] G Mintsis S Basbas P Papaioannou C Taxiltaris and I NTziavos ldquoApplications of GPS technology in the land trans-portation systemrdquo European Journal of Operational Researchvol 152 no 2 pp 399ndash409 2004

[7] NNandan ldquoOnline grid-based dynamic arrival time predictionusing GPS locationsrdquo International Journal of Machine Learningand Computing vol 3 no 6 pp 516ndash519 2013

[8] J Lu andG Chen ldquoA time-varying complex dynamical networkmodel and its controlled synchronization criteriardquo IEEE Trans-actions on Automatic Control vol 50 no 6 pp 841ndash846 2005

[9] J Lu X Yu G Chen and D Cheng ldquoCharacterizing thesynchronizability of small-world dynamical networksrdquo IEEETransactions on Circuits and Systems I Regular Papers vol 51no 4 pp 787ndash796 2004

[10] S Tan and J Lu ldquoCharacterizing the effect of populationheterogeneity on evolutionary dynamics on complex networksrdquoScientific Reports vol 4 article 5034 2014

[11] Y Chen J Lu X Yu and Z Lin ldquoConsensus of discrete-timesecond-order multiagent systems based on infinite productsof general stochastic matricesrdquo SIAM Journal on Control andOptimization vol 51 no 4 pp 3274ndash3301 2013

[12] S-H Chung A H I Lee and W L Pearn ldquoAnalytic net-work process (ANP) approach for product mix planning insemiconductor fabricatorrdquo International Journal of ProductionEconomics vol 96 no 1 pp 15ndash36 2005

[13] Y C Hu and Y L Liao ldquoUtilizing analytic hierarchy processto analyze consumersrsquo purchase evaluation factors of smart-phonesrdquoWorldAcademy of Science Engineering andTechnologyvol 78 pp 1047ndash1052 2013

[14] Y C Hu ldquoAnalytic network process for pattern classificationproblems using genetic algorithmsrdquo Information Sciences vol180 no 13 pp 2528ndash2539 2010


[15] Y C Hu J H Wang and R Y Wang ldquoEvaluating the perfor-mance of Taiwan Homestay using analytic network ProcessrdquoMathematical Problems in Engineering vol 2012 Article ID827193 24 pages 2012

[16] Y C Hu J H Wang and L P Hung ldquoEvaluating the e-servicequality of microbloggingrdquo in Proceedings of the InternationalSymposium on the Analytic Hierarchy Process Naples Italy 2011

[17] C-L Lin M-S Hsieh and G-H Tzeng ldquoEvaluating VehicleTelematics System by using a novel MCDM techniques withdependence and feedbackrdquo Expert Systems with Applicationsvol 37 no 10 pp 6723ndash6736 2010

[18] W-W Wu ldquoChoosing knowledge management strategies byusing a combined ANP and DEMATEL approachrdquo ExpertSystems with Applications vol 35 no 3 pp 828ndash835 2008

[19] J L Yang and G-H Tzeng ldquoAn integrated MCDM techniquecombined with DEMATEL for a novel cluster-weighted withANP methodrdquo Expert Systems with Applications vol 38 no 3pp 1417ndash1424 2011

[20] G-H Tzeng and J-J Huang Multiple Attribute Decision Mak-ing Methods and Applications CRC Press Boca Raton FlaUSA 2011

[21] C Y Hern ldquoSchedule planning for the development of intelli-gent transportation systems (ITS) in Taiwan areardquo Transporta-tion Planning Journal vol 29 no 1 pp 109ndash142 2000

[22] Y J Chiu and G H Tzeng ldquoEvaluating intelligent trans-portation security systems using MCDMrdquo in Proceedings ofthe 30th International Conference on Computers and IndustrialEngineering pp 131ndash136 Tinos Island Greece June-July 2002

[23] B K S Cheung K L Choy C L Li W Shi and J TangldquoDynamic routing model and solution methods for fleet man-agement with mobile technologiesrdquo International Journal ofProduction Economics vol 113 no 2 pp 694ndash705 2008

[24] E E Adam and R J Ebert Production and Operations Manage-ment ConceptsModels and Behaviour PrenticeHall NewYorkNY USA 5th edition 1991

[25] Definition of Global Positioning Systems The American HeritageDictionary Houghton Mifflin Boston Mass USA 4th edition2000

[26] C R Drane and C Rizos Positioning Systems in IntelligentTransportation Systems Artech House Publishers 1998

[27] Y ZhaoVehicle Location andNavigation Systems ArtechHousePublishers Norwood Mass USA 1997

[28] ATheiss D C Yen and C-Y Ku ldquoGlobal positioning systemsan analysis of applications current development and futureimplementationsrdquo Computer Standards and Interfaces vol 27no 2 pp 89ndash100 2005

[29] J Karp ldquoGPS in interstate trucking in Australia intelligencesurveillance- or compliance toolrdquo IEEE Technology and SocietyMagazine vol 33 no 2 pp 47ndash52 2014

[30] H Auernhammer ldquoPrecision farmingmdashthe environmentalchallengerdquoComputers and Electronics in Agriculture vol 30 no1ndash3 pp 31ndash43 2001

[31] Y P O Yang H M Shieh J D Leu and G H Tzeng ldquoA novelhybrid MCDM model combined with DEMATEL and ANPwith applicationsrdquo International Journal of Operations Researchvol 5 no 3 pp 160ndash168 2008

[32] Y-C Hu and J-F Tsai ldquoBackpropagation multi-layer percep-tron for incomplete pairwise comparison matrices in analytichierarchy processrdquo Applied Mathematics and Computation vol180 no 1 pp 53ndash62 2006

[33] Z Xu and C Wei ldquoConsistency improving method in theanalytic hierarchy processrdquo European Journal of OperationalResearch vol 116 no 2 pp 443ndash449 1999

[34] J A Martilla and J C James ldquoImportance-performance analy-sisrdquo Journal of Marketing vol 41 no 1 pp 77ndash79 1977

[35] C C ChenK C Chen and J R Chen ldquoThe study of key successfactors of ERP implementation in the small businessrdquo Journal ofChinese Economic Research vol 10 no 2 pp 31ndash42 2012

[36] H Y Chiou Analyses of the critical success factors on theimplementation of ERP system a study in the point of ERP projectmanager [Master thesis] Shih Chien University Taipei Taiwan2010

[37] J H HuangApply analytic network process to explore the criticalsuccess factors for enterprises implementing ERP systems [MSthesis] National Kaohsiung University of Applied SciencesKaohsiung Taiwan 2012

[38] S M Huang S I Chang and K H Su ldquoCritical success factorsfor implementing BS7799 information security managementsystem-based on petrochemical industryrdquo Journal of Informa-tion Management vol 13 no 2 pp 171ndash192 2006

[39] H C LeeApplying grey analytic hierarchy process to analyze thecritical success factors of ERP [MS thesis] Huafan UniversityTaipei Taiwan 2007

[40] H C Lin Exploration of key successful factors of ERP implemen-tation for small and medium firms [MS thesis] National ChengKung University Tainan Taiwan 2010

[41] C M Liu Critical success factors research of information systemof military organization implementation example of army train-ing and supply systems [MS thesis] Southern TaiwanUniversityof Science and Technology Tainan Taiwan 2012

[42] J C Pai G G Lee W G Tseng and Y L Chang ldquoOrga-nizational technological and environmental factors affectingthe implementation of ERP systems multiple-case study inTaiwanrdquo Journal of Electronic Commerce Studies vol 5 no 2pp 175ndash195 2007

[43] I H Sheu Influence enterprise resources plan system CSF(Critical Success Factor) implement successmdashfrom consultantdiscussion viewpoint [MS thesis] National Kaohsiung FirstUniversity Kaohsiung Taiwan 2006

Research ArticleImage-Based Pothole Detection System for ITS Serviceand Road Management System

Seung-Ki Ryu1 Taehyeong Kim1 and Young-Ro Kim2

1Highway and Transportation Research Institute Korea Institute of Civil Engineering and Building Technology283 Goyangdae-ro Ilsanseo-gu Goyang-si 411-712 Republic of Korea2Department of Computer Science and Information Myongji College Seoul 120-848 Republic of Korea

Correspondence should be addressed to Taehyeong Kim tommykimkictrekr



Copyright copy 2015 Seung-Ki Ryu et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Potholes can generate damage such as flat tire and wheel damage impact and damage of lower vehicle vehicle collision andmajor accidents Thus accurately and quickly detecting potholes is one of the important tasks for determining proper strategiesin ITS (Intelligent Transportation System) service and road management system Several efforts have been made for developinga technology which can automatically detect and recognize potholes In this study a pothole detection method based on two-dimensional (2D) images is proposed for improving the existing method and designing a pothole detection system to be appliedto ITS service and road management system For experiments 2D road images that were collected by a survey vehicle in Koreawere used and the performance of the proposed method was compared with that of the existing method for several conditionssuch as road recording and brightness The results are promising and the information extracted using the proposed method canbe used not only in determining the preliminary maintenance for a road management system and in taking immediate action fortheir repair and maintenance but also in providing alert information of potholes to drivers as one of ITS services

1 Introduction

Apothole is defined as a bowl-shaped depression in the pave-ment surface and its minimum plan dimension is 150mm[1] With the climate change such as heavy rains and snow inKorea damaged pavements like potholes are increasing andthus complaints and lawsuits of accidents related to potholesare growingThere are internal causes to potholes such as thedegradation and responsiveness or durability of the pavementmaterial itself to climate change such as heavy rainfall andsnowfall and external causes such as the lack of qualitymanagement and construction management

Also Table 1 shows the number of compensations andcompensation amounts about accidents related to road facil-ities for 6 years 2008 to 2013 in Seoul [2]

As shown in Table 1 the number of compensations andcompensation amounts related to potholes occupymore than50 of total the number of compensations and compensationamounts in Seoul city Seoul city has been pouring attention

to prepare a countermeasure of potholes that threaten roadsafety in this way

As one type of pavement distresses potholes are impor-tant clues that indicate the structural defects of the asphaltroad and accurately detecting these potholes is an importanttask in determining the proper strategies of asphalt-surfacedpavement maintenance and rehabilitation However manu-ally detecting and evaluatingmethods are expensive and timeconsumingThus several efforts have beenmade for develop-ing a technology that can automatically detect and recognizepotholes whichmay contribute to the improvement in surveyefficiency and pavement quality through prior investigationand immediate action

Existing methods for pothole detection can be dividedinto vibration-based methods three-dimensional (3D) re-construction-based methods and vision-based methods [3ndash26] Although these vision-based methods are cost-effectivecompared with 3D laser scanner methods it may be difficultto accurately detect a pothole using these methods because of



Table 1The number of compensations and compensation amountsabout accidents for 6 years (2008 to 2013) in Seoul

Classification Total accidents Pothole related Rate ()The number ofcompensations 2471 1745 706

Compensationamounts ($) 4440000 2370000 534

the distorted signals generated by noise in collecting imageand video data Thus a pothole detection method usingvarious features in 2D images is proposed for improvingthe existing pothole detection method [20] and accuratelydetecting a pothole Also the performance of the proposedmethod is compared with that of the existing method forseveral conditions such as road recording and brightnessFurther a pothole detection system is designed to be appliedto ITS service and road management system The designedand developed pothole detection system is expected to beused not only in determining the preliminary maintenanceof road management system and in taking immediate actionfor their repair and maintenance but also in providing alertinformation of potholes to drivers as one of ITS services

2 Literature Review

Several efforts have been made for developing a methodwhich can automatically detect and recognize potholesDetailed surveys on methods for pothole detection can befound in Koch and Brilakis [20] and Kim and Ryu [27]Existing methods for pothole detection can be divided intovibration-based methods by B X Yu and X Yu [3] De Zoysaet al [4] Eriksson et al [5] and Mednis et al [6] three-dimensional (3D) reconstruction-based methods by Wang[7] Kelvin [8] Chang et al [9] Vijay [10] Hou et al [11] Li etal [12] Salari et al [13] Staniek [14] Zhang et al [15] Joubertet al [16] andMoazzam et al [17] and vision-basedmethodsby Wang and Gong [18] Lin and Liu [19] Koch and Brilakis[20] Jog et al [21] Huidrom et al [22] Koch et al [23] Buzaet al [24] Lokeshwor et al [25] and Kim and Ryu [26]

Vibration-based method uses accelerometers in order todetect potholes Considering the advantages of a vibration-based system these methods require small storage and canbe used in real-time processing However vibration-basedmethods could provide the wrong results for example thatthe hinges and joints on the road can be detected as potholesand that potholes in the center of a lane cannot be detectedusing accelerometers due to not being hit by any of thevehiclersquos wheels (Eriksson et al) [5]

3D laser scanner methods can detect potholes in realtime However the cost of laser scanning equipment is stillsignificant at the vehicle level and currently these works arefocused on the accuracy of 3D measurement Stereo visionmethods need a high computational effort to reconstructpavement surfaces through matching feature points betweentwo views so that it is difficult to use them in a real-timeenvironment [7 8 10 11 13ndash15] Recently Moazzam et al [17]used a low-cost Kinect sensor to collect the pavement depth

images and calculate the approximate volume of a potholeAlthough it is cost-effective as compared with industrialcameras and lasers the use of infrared technology based ona Kinect sensor for measurement is still a novel idea andfurther research is necessary for improvement in error rates

A 2D image-based approach has been focused only onpothole detection and is limited to a single frame so itcannot determine the magnitude of potholes for assessmentTo overcome the limitation of the abovemethod video-basedapproaches were proposed to detect a pothole and calculatethe total number of potholes over a sequence of frames

Although these vision-based methods are cost-effectivecompared with 3D laser scanner methods it may be difficultto accurately detect a pothole using these methods because ofthe distorted signals generated by noise in collecting imageand video data Thus a pothole detection method usingvarious features in 2D images is proposed for improvingthe existing pothole detection method [20] and accuratelydetecting a pothole Also the performance of the proposedmethod is compared with that of the existing method forseveral conditions such as road recording and brightness Inour study for comparison the method by Koch and Brilakis[20] was selected because their method had a good result ascompared to other existing methods

3 The Pothole Detection System

A pothole detection system was designed to collect roadimages through a newly developed optical devicemounted ona vehicle and detects a pothole from the collected data usingthe proposed algorithm Figure 1 shows a pothole detectionsystem that was developed in this study and its applicationThis system includes an optical device and a pothole detectionalgorithm

The optical device on a vehicle collects potholes data andthe collected data is sent to a pothole detection algorithmAlso the pothole information such as the location andseverity of a pothole obtained from a pothole detectionalgorithm is sent to a road management center The opticaldevice was designed to easily be mounted in a vehicle and ithas several functions such as collecting and storing data ofpotholes communicating by Wi-Fi and gathering locationinformation by GPS Table 2 shows the detailed specificationof the optical device

The pothole information obtained from a pothole detec-tion system is sent to a center and can be applied to a potholealert service and the road management system As shownin Figure 2 pothole information is sent from a center toRSEs (Roadside Equipment) and navigation companies andthen the information is sent to OBUs (Onboard Unit) ornavigations through DSRC (Dedicated Short-Range Com-munication) and WAVE communication Finally potholealert information such as location and severity is displayed onOBU or navigation Before passing the pothole a driver canrecognize the presence of the pothole in advance and avoidrisks Pothole alert service is still a novel idea and furtherresearch is necessary for improvement in image processingtime and communication


Potholeimages

Pothole information(location and severity)

Vehicle stationary

Pothole detectionalgorithm

optics

Center

Pothole alert service

Road managementsystem

PPPP tP tPotPotPotPoth lh lh lholholholhol ddde de de de d tteteeteeteete iititictictictictionononon

Figure 1 Pothole detection system and its application

Center

RSE

company

OBU

NavigationNavigation

Pothole information

Potholeinformation

Driver and carThrough DSRC

or WAVE

Through Wi-Fi or LTE

Display of pothole alert information(location and

severity)

or

Figure 2 Pothole alert service

Table 2 Specification of the optical device [26]

Item SpecificationHousing (i) PlasticSize (i) 110 (119882) lowast 40 (119871) lowast 110 (119867)Range (i) The inside lane left and right lanesResolution (i) 1280 lowast 720 60 fps

Camera module (i) 6 glasses and CMOS fixed type(ii) The diameter of lenses 12mm

CPU (i) More than 3000DMIPSStorage (i) Two storage spaces for safety

GPS (i) Antenna 25mm (119882) times 25mm (119871)(ii) Backup battery

Power (i) Using the power of a vehicle(ii) Holding secondary power unit

Also the obtained pothole information is provided tothe Road Management System and the repair time andmaintenance quantities are determined according to theseverity and location of the pothole

4 The Proposed Pothole Detection Method

The proposed method can be divided into three steps (1)segmentation (2) candidate region extraction and (3) deci-sion (Figure 3) First a histogram and the closing operation

of a morphology filter are used for extracting dark regions forpothole detection Next candidate regions of a pothole areextracted using various features such as size and compact-ness Finally a decision is made whether candidate regionsare potholes or not by comparing pothole and backgroundfeatures

The segmentation step is to separate a pothole regionfrom the background region by transforming an originalcolor image into a binary image using the histogram of aninput image HST (Histogram Shape-Based Thresholding)maximum entropy and Otsu [28] can be used for thistransformation into a binary image In this study an inputimage is transformed into a binary image using HST [20]

The candidate step involves extracting a pothole candi-date region from a binary image obtained in the segmentationstep First the median filter is used to remove noise such ascracks and spots 3 times 3 7 times 7 and 9 times 9 filters were tested andthe 9 times 9 filter showed the best performance among the threefilters

Next the damaged outlines of object regions are restoredand small pieces are removed using the closing operation(dilation and erosion) of a morphology filter A square (7 times7) type of the structure element was used for the closingoperation


Segmentation Candidate Decision

Input image

Binarization by HST

Segmented images

Morphologyoperation (closing)

Feature basedcandidate extraction

Candidaterefinement

Ordered histogram intersection

Pothole decision(OHI Sobel)

Detected pothole region

Candidate region

Noise filtering(median filter)

Figure 3 Process of the proposed pothole detection method

After the closing operation candidate regions are ex-tracted using features such as size compactness ellipticityand linearity as shown in

119862V

=

1 if 119878 (1198721015840119888) gt 119879119904 Com (1198721015840

119888) gt 119879com and so forth

0 otherwise

(1)

where119862V the value of region119862 for the candidate in the image119878(1198721015840

119888) the size of region 119862 in the image after the closing

operation Com(1198721015840119888) the compactness of region 119862 in the

image after the closing operation 119879119904 the threshold for size

and 119879com the threshold for compactness

The size of a region 119862 is defined as total number of pixelsin the region119862which depends on a size of a pothole an objectdistance and a focal length Also compactness is defined as

com (1198721015840119888) =1198972

4120587119860 (2)

where 119897 the perimeter and 119860 the area of region 119862Also the refinement of candidate regions is needed

to detect the correct pothole regions after obtaining thecandidate regions The initial candidates obtained usingfeatures may not represent the full-sized pothole area Thusthe refinement of candidate regions using features such ascompactness center point and convex hull is necessarybefore it can be decided whether various and incompletecandidate regions such as shades spots and patches arepotholes or not Incomplete candidate regions are refinedusing the convex hull operation according to the decision of

1198621015840

V =

result of convex hull operation if 119862119888isin 119862 Com (119862) gt 119879com and so forth

119862V otherwise(3)

where 1198621015840V the value of refined region 1198621015840 for the candidatein the image 119862V the value of region 119862 for the candidate inthe image 119862

119888 the center position of region 119862 Com(119862) the

compactness of region119862 in the image and119879com the thresholdfor compactness

Next MHST (modified HST) separates not only thepothole region but also a bright region such as a lanemarking from the background region

The decision step involves deciding whether the refinedcandidate regions are potholes or not after the comparison ofcandidate regions with the background region using featuressuch as standard deviation and histogram

In particular as a histogram feature ordered histogramintersection (OHI) [29] is used in this study By using OHIit is possible to distinguish stains patches light shades

and so forth that cannot be separated from potholes usingthe existing method [20] and to avoid the wrong detectionof potholes OHI is a method of measuring the degreeof similarity between regions in an image Although someproblems occur with noise or when there is a change inbrightness OHI can measure the degree of similarity byidentifying these differences OHI can be expressed as shownin

OHI (ℎ119888 ℎ119887) =

119899

sum

119894=0

min (oh119894119888 oh119894119887) (4)

where OHI(ℎ119888 ℎ119887) OHI for candidate region 119888 and back-

ground region 119887 oh119894119888 the ordered histogram of index 119894 for

candidate region 119888 oh119894119887 the ordered histogram of index 119894 for

background region 119887 119894 the index of histogram (119894 = 0 to 255


for 8 bits) and 119899 themaximumnumber of the index (119899 = 255for 8 bits)

In this study if the standard deviation of the refinedcandidate region is smaller than the threshold for standarddeviation (119879std) or if OHI of the pixels between the refined

candidate region and the background region is close to 1 andthe OHI of values using the Sobel operation [30] is close to 1it is decided that the refined candidate region is not a potholebecause it is similar to the background region Equation (5)shows this discriminant

119901

=

non-pothole region if Std1198881015840 lt 119879std or (OHI (ℎ

1198881015840 ℎ119887) gt 119879119900 OHI (ℎ1015840

1198881015840 ℎ1015840

119887) gt 1198791199001015840) (Outregionstd minus Innerregionstd) lt 119879std1015840 (Outregionave minus Innerregionave) gt 119879ave

pothole region otherwise

(5)

where Std1198881015840 the standard deviation of the refined candidate

region 1198881015840 OHI(ℎ1198881015840 ℎ119887) OHI for the refined candidate region

1198881015840 and background region 119887 OHI(ℎ1015840

1198881015840 ℎ1015840

119887) OHI for the refined

candidate region 1198881015840 and background region 119887 using theSobel operation Outregionstd the standard deviation of theoutside of the refined candidate region Innerregionstd thestandard deviation of the inside of the refined candidateregion Outregionave the average of the outside of the refinedcandidate region Innerregionave the average of the inside ofthe refined candidate region 119879std the threshold for standarddeviation119879std1015840 the threshold for standard deviation of valuesby the Sobel operation 119879ave the threshold for average 119879119900 thethreshold for OHI and 119879

1199001015840 the threshold for OHI of values

by the Sobel operationFigure 4 shows the result image at each step by the

proposed method

5 Experiment Results

In this study 2D road images that had been collected bya survey vehicle in Korea from May to June 2014 wereused Two-dimensional images with a pothole and without apothole extracted from the collected pothole database (a totalof 150 video clips) were used to compare the performance ofthe proposed method with that of the existing method [20]by several conditions such as road recording and brightnessconditions

To collect video data of potholes the newly developedoptical device (resolution 1280 times 720 60 fs) were mountedat the height of a rear-view mirror of a survey vehicle andthey recorded the road surfaces ahead during movement

The proposed pothole detection method was imple-mented in Microsoft Visual C++ 60 The image processingwas performed on a laptop (Intel Core i5-4210U 24GHz8GB RAM) Table 3 shows the values of thresholds used inthis study All threshold values except for 119879

ℎ(threshold for

HST and MHST) were empirically set as the most suitablevalue to distinguish various types of potholes from similarobjects

A total of 90 images were randomly chosen from 100video clips for experiments For experiments by road condi-tion 20 asphalt images and 20 concrete images were selectedrandomly and Figure 5 shows the examples and results of theselected images for experiment by road condition

Table 3 The values of thresholds used in this study

Thresholds Values Thresholds Values

119879ℎ

The valuedepends on the

image119879std1015840 10

119879119904 512 119879ave 00119879com 005 119879

119900087

119879std 8 1198791199001015840 085

In Figure 5 it is shown that the proposed methodaccurately detects most of the potholes in both asphalt andconcrete images Fourth image from the left among asphaltimages has stains and the proposed method does not detectthem as potholes but the existing method [20] detects themas potholes

For experiments by recording condition 10 originalimages and 10 images by close-up were selected and Figure 6shows the examples and results of the selected images forexperiment by recording condition

In Figure 6 it is shown that the proposed method accu-rately detects most of the potholes in close-up images A fewresults show that only a portion of the pothole was detectedbecause only that part of the pothole was extracted as acandidate region

Also for experiments by brightness condition 10 brightimages (average gray level gt 120) and 10 dark images (averagegray level lt 110) were selected and Figure 7 shows theexamples and results of the selected images for experimentby brightness condition

The proposedmethod has a better performance for brightimages rather than dark images Not only the proposedmethod but also all existing methods detect dark regions assuspected potholes Thus it is obvious that the performanceof detecting potholes under dark circumstances is worse thanthat of detecting potholes under normal brightness

In addition 30 more images for experiments were testedand the result of pothole detection of experiments usingthe proposed method and existing method for a total of90 images are summarized in Table 4 In order to comparethe performance of the proposed method with that of theexisting method [20] image segmentation and candidateextraction were processed under the same conditions andthe decision criteria for a pothole were applied differently


(1) Original (2) HST (3) Inversion (4) Median filter

(5) Dilation (6) Erosion (7) Candidate (8) Refinement

(9) Sobel (10) Erosion (11) Edge (12) Decision

Figure 4 Result images at each step using the proposed method

according to the proposed criteria in each method In thistable in order to represent accurate detection performancethe number of true positives (TP correctly detected as apothole) false positives (FP wrongly detected as a pothole)true negatives (TN correctly detected as a nonpothole) andfalse negatives (FN wrongly detected as a nonpothole) [19]was counted manually Also accuracy precision and recallusing the proposed method and the existing method werecalculated as measurements for validation

(1) accuracy the average correctness of a classificationprocess minus (TP + TN)(TP + FP + TN + FN)

(2) precision the ratio of correctly detected potholes tothe total number of detected potholesminusTP(TP+FP)

(3) recall the ratio of correctly detected potholes to actualpotholes minus TP(TP + FN)

In our study for comparison the method by Koch andBrilakis [20] was selected because their method had a goodresult as compared to other existing methods Table 4 showsthat the proposed method reaches an overall accuracy of735 with 800 precision and 733 recall All threemeasures validate that most potholes in images can be

Table 4 Performance comparison

Performances The existing method The proposed methodTotal TP 22 44Total FP 18 11Total TN 24 31Total FN 38 16Accuracy 451 735Precision 550 800Recall 367 733

correctly detected Also the results of the proposed methodshow a much better performance than that of the existingmethod which has an overall accuracy of 451 with 550precision and 367 recall By the existing method it isdifficult to separate stains or patches similar to a potholefrom an actual pothole using only the feature of standarddeviation However the proposed method can accuratelydetect a pothole using several features such as the standarddeviation of a candidate region OHI differences in thestandard deviations and averages between the outside andinside of a candidate region It is shown that a joint part


(a) Asphalt images

(b) Concrete images

Figure 5 Examples and results of the selected images for road condition

between an asphalt road and a concrete road was incorrectlydetected However this wrong detection can be removed laterby adding a feature corresponding to the concrete in thedecision step

Also the processing times for the proposed method werecalculated through 10 of images that were chosen randomlyTable 5 shows the calculated processing times for the pro-posed method It is impossible to compare the processingtimes of the proposedmethodwith those ofKoch andBrilakis[20] exactly since it is impossible to implement Koch andBrilakisrsquo method in their same experiment circumstance andit can result in needing more times for the Koch and Brilakisrsquomethod due to the wrong setting for experiments Howeverthe processing times of the Koch and Brilakisrsquo method can bereferred to Koch et al [23]

Table 5 shows that more processing times are needed forImages 3 7 and 8 since those images have more numbersof candidate regions or bigger regions than other images It

is obvious that the proposed method needs more processingtime than Koch and Brilakis [20] because the proposedmethod uses various features for detecting potholes Furtherwork for improving image processing time is necessary forthe pothole detection system to be applied to real-time pot-hole detection and real pothole alert service

The results are promising and the information extractedusing the proposed method can be used not only in deter-mining the preliminary maintenance for a road managementsystem and in taking immediate action for their repair andmaintenance but also in providing alert information ofpotholes to drivers as one of ITS services

6 Conclusions

In this study a pothole detection method based on 2D roadimages was proposed for improving the existing methodand designing a pothole detection system to be applied to


Table 5 Processing times

Images Segmentation (sec) Candidate (sec) Decision (sec) Total (sec)1 65 146 04 2152 65 174 04 2433 63 611 04 6784 68 177 04 2495 63 192 04 2596 63 85 04 1527 63 343 04 4108 63 83 03 1499 70 2107 05 218210 63 70 04 137Average 65 399 04 468

(a) Original images

(b) Close-up images

Figure 6 Examples and results of the selected images for recording condition


(a) Bright images

(b) Dark images

Figure 7 Examples and results of the selected images for brightness condition

ITS service and road management system For experiments2D road images that were collected by a survey vehiclein Korea were used and the performance of the proposedmethod was compared with that of the existing method forseveral conditions such as road recording and brightnessRegarding the experiment results the proposed methodreaches an overall accuracy of 735 with 800 precisionand 733 recall which is a much better performance thanthat of the existing method having an overall accuracy of451 with 550 precision and 367 recall

However there are some limitations in the proposedmethod Potholes may be falsely detected according to thetype of shadow and various shapes of potholes Thus inorder to more accurately detect potholes it is necessary touse images from not only a single sensor but also additionalsensors and to add to the proposed method more featuresfor these sensors Also the stability of the pothole detection

method based on two-dimensional images needs to be addedbecause the vehiclersquos vibration during driving will have bigaffection on the detecting equipment The proposed methodwill have a more improved performance through moreexperiments under a variety of circumstances In additionthe proposed method needs more processing time than Kochand Brilakis [20] because the proposed method uses variousfeatures for detecting potholes Therefore further work forimproving image processing time and performance of theproposed method is necessary for the pothole detectionsystem to be applied to real-time pothole detection and realpothole alert service




Acknowledgment

This research was supported by a grant from a StrategicResearch Project (Development of Pothole-Free Smart Qual-ity Terminal [2014-0219]) funded by the Korea Institute ofCivil Engineering and Building Technology

References

[1] J S Miller and W Y Bellinger ldquoDistress identification manualfor the long-term pavement performance programrdquo FHWARD-03-031 Federal HighwayAdministrationWashington DCUSA 2003

[2] MOLIT (Ministry of Land and Infrastructure and Transport inKorea) Data for Inspection of Government Agencies 2013

[3] B X Yu and X Yu ldquoVibration-based system for pavementcondition evaluationrdquo in Proceedings of the 9th InternationalConference on Applications of Advanced Technology in Trans-portation pp 183ndash189 August 2006

[4] K De Zoysa C Keppitiyagama G P Seneviratne and WW A T Shihan ldquoA public transport system based sensornetwork for road surface condition monitoringrdquo in Proceedingsof the 1st ACM SIGCOMMWorkshop on Networked Systems forDeveloping Regions (NSDR 07) Tokyo Japan August 2007

[5] J Eriksson L Girod B Hull R Newton S Madden andH Balakrishnan ldquoThe pothole patrol using a mobile sensornetwork for road surface monitoringrdquo in Proceedings of the 6thInternational Conference on Mobile Systems Applications andServices (MobiSys rsquo08) pp 29ndash39 June 2008

[6] AMednis G Strazdins R Zviedris G Kanonirs and L SelavoldquoReal time pothole detection using Android smartphones withaccelerometersrdquo in Proceedings of the International Conferenceon Distributed Computing in Sensor Systems and Workshops(DCOSS rsquo11) pp 1ndash6 IEEE Barcelona Spain June 2011

[7] K C P Wang ldquoChallenges and feasibility for comprehensiveautomated survey of pavement conditionsrdquo in Proceedings ofthe 8th International Conference on Applications of AdvancedTechnologies in Transportaion Engineering pp 531ndash536 May2004

[8] C P Kelvin ldquoAutomated pavement distress survey throughstereovisionrdquo Technical Report of Highway IDEA Project 88Transportation Research Board 2004

[9] K T Chang J R Chang and J K Liu ldquoDetection of pavementdistresses using 3D laser scanning technologyrdquo in Proceedingsof the ASCE International Conference on Computing in CivilEngineering pp 1085ndash1095 July 2005

[10] S Vijay Low costmdashFPGA based system for pothole detection onIndian roads [MS thesis of Technology] Kanwal Rekhi Schoolof Information Technology Indian Institute of TechnologyMumbai India 2006

[11] Z Hou K C P Wang and W Gong ldquoExperimentation of 3Dpavement imaging through stereovisionrdquo in Proceedings of theInternational Conference on Transportation Engineering (ICTErsquo07) pp 376ndash381 Chengdu China July 2007

[12] Q Li M Yao X Yao and B Xu ldquoA real-time 3D scanning sys-tem for pavement distortion inspectionrdquo Measurement Scienceand Technology vol 21 no 1 Article ID 015702 2010

[13] E Salari E Chou and J Lynch ldquoPavement distress evalua-tion using 3D depth information from stereo visionrdquo TechRep MIOH UTC TS43 2012-Final Michigan-Ohio UniversityTransporation Center 2012

[14] M Staniek ldquoStereo vision techniques in the road pavementevaluationrdquo in Proceedings of the 28th International Baltic RoadConference pp 1ndash5 Vilnius Lituania August 2013

[15] Z Zhang XAi C KChan andNDahnoun ldquoAn efficient algo-rithm for pothole detection using stereo visionrdquo in Proceedingsof the IEEE International Conference on Acoustics Speech andSignal Processing (ICASSP rsquo14) pp 564ndash568 Florence ItalyMay2014

[16] D Joubert A Tyatyantsi J Mphahlehle and V ManchidildquoPothole tagging systemrdquo in Proceedings of the 4th Robotics andMechanics Conference of South Africa pp 1ndash4 2011

[17] IMoazzamK Kamal SMathavan S Usman andMRahmanldquoMetrology and visualization of potholes using the microsoftkinect sensorrdquo in Proceedings of the 16th International IEEEConference on Intelligent Transportation Systems IntelligentTransportation Systems for All Modes (ITSC rsquo13) pp 1284ndash1291October 2013

[18] K C P Wang and W Gong ldquoReal-time automated surveysystem of pavement cracking in parallel environmentrdquo Journalof Infrastructure Systems vol 11 no 3 pp 154ndash164 2005

[19] J Lin and Y Liu ldquoPotholes detection based on SVM in thepavement distress imagerdquo in Proceedings of the 9th InternationalSymposium on Distributed Computing and Applications to Busi-ness Engineering and Science (DCABES 10) pp 544ndash547 HongKong China August 2010

[20] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011

[21] GM Jog C KochM Golparvar-Fard and I Brilakis ldquoPotholeproperties measurement through visual 2D recognition and3D reconstructionrdquo in Proceedings of the ASCE InternationalConference onComputing inCivil Engineering pp 553ndash560 June2012

[22] L Huidrom L K Das and S Sud ldquoMethod for automatedassessment of potholes cracks and patches from road surfacevideo clipsrdquo ProcediamdashSocial and Behavioral Sciences vol 104pp 312ndash321 2013

[23] C Koch G M Jog and I Brilakis ldquoAutomated pothole distressassessment using asphalt pavement video datardquo Journal ofComputing in Civil Engineering vol 27 no 4 pp 370ndash378 2013

[24] E Buza S Omanovic and A Huseinnovic ldquoPothole detectionwith image processing and spectral clusteringrdquo in Proceedingsof the 2nd International Conference on Information Technologyand Computer Networks pp 48ndash53 2013

[25] H Lokeshwor L K Das and S Goel ldquoRobust method forautomated segmentation of frames withwithout distress fromroad surface video clipsrdquo Journal of Transportation Engineeringvol 140 no 1 pp 31ndash41 2014

[26] T Kim and S Ryu ldquoSystem and method for detecting potholesbased on video datardquo Journal of Emerging Trends in Computingand Information Sciences vol 5 no 9 pp 703ndash709 2014

[27] T Kim and S Ryu ldquoReview and analysis of pothole detectionmethodsrdquo Journal of Emerging Trends in Computing and Infor-mation Sciences vol 5 no 8 pp 603ndash608 2014

[28] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979

[29] D V D Weken M Nachtegael and E E Kerre ldquoSome newsimilarity measures for histogramsrdquo in Proceedings of the 4thIndian Conference on Computer Vision Graphics amp ImageProcessing (ICVGIP rsquo04) Kolkata India 2004

[30] R Gonzalez and R Woods Digital Image Processing AddisonWesley Boston Mass USA 1992










Editorial Board










Contents















1 Introduction


















References
















1 Introduction













1198881 1198882 119888


1 1198882 119888

119899are


1 V2 V

119896 in



Custo

mer

-veh

icle

assig

nmen

t

Opt

imiz

ed as

signm

ent

CV

c1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

CV

c1c2

cn

12

k

middot



Fitness = min119899

sum

119894=0

119899

sum

119895=0

119896

sum

V=1119889119894119895119883

V119894119895+ 1198891198990119883

V1198990

119894 = 119895 (1)

119899

sum

119894=0

119899

sum

119895=0

119883

V119894119895119903119894le 119876V 119894 = 119895 V isin 119881 (2)

119883

V119894119895

=


0 otherwise

(3)


119894119895is the









1 1198882 119888

119899 to










119894= 119883

1198941 119883

119894119871





119894119871 whereas


119871


119881

new119894119895


times (119866119895minus 119883119894119895)

119883

new119894119895= 119883119894119895+ 119881

new119894119895

(4)



1

Depot

310

8

2

95

7

6

4

Opt

imiz

ed sc

hedu

le

Opt

imiz

ed as

signm

ent

1

Depot

72

8

10

95

3

6

4

7

Depot

310

8

5

92

1

6

4

CV

c1c2

cn

12

k

middot




119883

119881

119894119883

119881

1198941119883

119881


119881

119894119899119883

119881

119894119899+1119883

119881

119894119899+2sdot sdot sdot 119883

119881

119894119899+119896minus1











119894= 119883

119881

1198941 119883

119881

119894119899 119883

119881

119894119899+119896minus1


1Cus2 Cus

119899Veh1Veh2 Veh

119896minus1) For

















119863

1 119883

119863

2) is shown



Key2 13 08 24 19 02 12 21

02 08 12 13 19 2 21 24Key


Vehicle assignment

Visit order

Veh 1

Veh1

Veh1 Veh2

Veh2

Cus1

Cus1

Cus1

Veh 2

Cus2

Cus2

Cus2

Veh 3

Cus3

Cus3

Cus3

Cus4

Cus4

Cus4

Cus5

Cus5

Cus5

Cus6

Cus6

Cus6

XiV

XiV



X119863 119883

119863

1119883

119863

2






119862 119910119862)) can be yielded by

119909119862=

sum

119896

119894=1119909

V119894+ 119909119874

119896 + 1

119910119862=

sum

119896

119894=1119910

V119894+ 119910119874

119896 + 1

(5)

F

O

DE

G

HA

I

C

J

B

K

Route-1

Route-2

Route-3


F

O

DE

G

HA

I

C

J

B

K

dE

dF

dD

dC

dB

dA

CC


In (5) 119909119862and 119910





119874are the




F

O

DE

B

C

JK

G

H

I

A

(a) 1198891 = 119874119861 + 119861119866minus 119874119866

F

O

DE

B

C

JK

G

H

I

A

(b) 1198892 = 119866119861 + 119861119867minus 119866119867

F

O

DE

C

J

A

K

G

H

IB

(c) 1198893 = 119867119861 + 119861119868 minus 119867119868

F

O

DE

B

C

J

A

K

G

H

I

(d) 1198894 = 119868119861 + 119861119874minus 119868119874








(3) Outer layer PSO



layer PSO





Start




End

Yes Yes

NoNo

YesNo


Initialize VVX

V

Update VVX

V

Initialize VDX

D

Update VDX

D

search(XV)

Fitness(X ) lt

Fitness(XV)

Update(SA)

Fitness( )

Updateand

Pass XD

to inner layer PSO

Fitness(XD) =

Fitness( )= XLSV

GVbest

XVnew

PVbest

XVnew X

Vnew

Updateand

GVbest

PVbest

GVbest

LSV

XVLS = local




119899









A-n54-k9 53 + 53 53 + 8

A-n64-k8 63 + 63 63 + 7







1= 2 and 119888

2= 1 and the


1= 1198882= 2 and 100



BKStimes 100 (6)


Avg Dev () =sum

119899

119894=1DEV119894

119899

(7)




1) RBI (LS

2)




0102030405060708090

Aver

age d

evia

tion

()

A-n3

2-k5

A-n3

3-k5

A-n3

6-k5

A-n4

5-k6

A-n4

5-k7

A-n5

5-k9

A-n6

0-k9

A-n6

2-k8

A-n6

4-k9

Aver

age

wo LSLS1

LS2LS3




Imp() =Fitness


Fitness119908119900

times 100 (8)





Fitn

ess

950

900

850

800

750

700

Iterations

0

50

100

150

200

250

300

350

400

450

500

550

600

650

700

750

800

850

900

950

1000

(a)

Fitn

ess

830

810

790

770

750

730

710

690

670

650

Iterations

0 5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

(b)


(a) (b)






(a) (b)



5 Conclusions






Acknowledgment


References




































1 Introduction



























HMM correction

Vehicle V1

Vehicle V2

Vehicle Vn



based on HMM

Entireroutes

Routerecommendation


Vehicle Vi

A set ofOutput

Input

RS(Tk Roadi)

RouteT119896



upcomingroutes


Rm1Rm2

Rm3

Rm4

Rm5

Rm1x

Rm2y

Rm3x Rm4x

Rm5y

Y

X0


Ri

Rj

Rk

Rix

Rjy

Rkx

Rky

Y

X0



119860119899

997888rarr 119861 = 1198771119909 (1198771119910)


(1)





= (1 3) (1 4) (3 4) (3 1)

(2)






Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd


Sichuan RdJinmao St

Longpan Rd

Jianning Rd

0 1 2 3 4 5 6 7 80

1

2

3

4

5

6

Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd

Central Rd

Xufu Alley

Sichuan Rd

Jinmao St

Longpan Rd

Jianning Rd

X

Y












1199051 lt (1 3) (1 4) (3 4) (3 1) gt

1199052 lt (3 1) (3 4) (1 4) (1 3) gt

1199053 lt (0 3) (1 3) (1 5) (4 5) gt

1199054 lt (0 3) (0 0) (0 4) (4 1) gt

1199055 lt (2 0) (2 1) (3 1) (3 2) (4 2) gt

1199051 lt (1 3) (1 4) (3 4) (3 1) gt

(3)














997888997888997888997888rarr119860119895119861119895 = (119909119895119899minus1199091198951 119910119895119899minus



cos ⟨ 119886119894 119886119895⟩ =

119886119894 sdot 119886119895

1003816100381610038161003816 1198861198941003816100381610038161003816 sdot10038161003816100381610038161003816119886119895

10038161003816100381610038161003816

= ((119909119894119899 minus 1199091198941) sdot (119910119894119899 minus 1199101198941)

+ (119909119895119899 minus 1199091198951) sdot (119910119895119899 minus 1199101198951))


2

sdotradic(119909119895119899 minus 1199091198951)2

+ (119910119895119899 minus 1199101198951)2

)

minus1

(4)


1





120587119894 =

Count (119904119863119894

) + 120582Count (119904New119863119894

)

sum119899

119895=1[Count (119904119863

119895

) + 120582Count (119904New119863119895

)]

(5)




1198961= 1199001198962)






119894

)

and Count(119904New119863119894



119875 (119904119894 | 119904119894minus1)

=

Count (119904119863119894minus1

119904119863119894

) + 120582Count (119904New119863119894minus1

119904New119863119894

) + 1

Count (119904119863119894minus1

) + 120582Count (119904New119863119894minus1

) + 119898

(6)


119904119863119894


119904New119863119894

)



119875 (V119895 | 119904119894)

=

Count (119904119863119894minus1

V119863119894

) + 120582Count (119904New119863119894minus1

VNew119863119894

) + 1

Count (119904119863119894

) + 120582Count (119904New119863119894

) + 119899

(7)


V119863119894


VNew119863119894

)






sequence

Viterbialgorithm










119901119894(1199091) 119901119894(119909

2) 119901119894(119909



1) 119901119894(119910

2) 119901119894(119910

119899) (1198632 notin 119881) and the











(b) Suppose that 1198632 = 119901119894(1199101) 119901119894(119910

2) 119901119894119896 then


(c) Suppose that 1198632 = 1199011198941 119901119894(1199102) 119901119894(119910

119899) then



1) 119901119894(119910

2) 119901119894(119910

119899) and 1199101 = 1 119910119899


1)

119901119894(1199102) 119901119894(119910





















119875119894 =sum119899

119896=1119863(119877119896 119862119877)

sum119899

119905=1Dist 1003816100381610038161003816119877119905

1003816100381610038161003816

times 100 (8)



1198751 =Repeat (1198771 1198771) + Repeat (1198773 1198771)

Dist 100381610038161003816100381611987711003816100381610038161003816 + Dist 10038161003816100381610038161198773

1003816100381610038161003816

times 100

=6 + 1

6 + 7times 100 = 5385

(9)












(a) (b)


100908070605040302010

01009080706050403020100

Trip completed ()

Cor

rect

pre

dict

ion

()


Cor

rect

pre

dict

ion

()

100908070605040302010

01009080706050403020100

Trip completed ()


Cor

rect

pre

dict

ion

()


100908070605040302010

0


120582 = 05

120582 = 02


Cor

rect

pre

dict

ion

()

100908070605040302010

0



120582 = 05

120582 = 02







90

80

70

60

50

40

30

20

10

0Other time periods

Cor

rect

pre

dict

ion

()

Time of day







7 Conclusion






Acknowledgments


References

























1 Introduction



















1198811198732

1198811198733

1198811198734

1198811198731

(1198811198731 1198811198731) (119881119873

1 1198811198732) (119881119873

1 1198811198733) (119881119873

1 1198811198734)

1198811198732

(1198811198732 1198811198731) (119881119873

2 1198811198732) (119881119873

2 1198811198733) (119881119873

2 1198811198734)

1198811198733

(1198811198733 1198811198731) (119881119873

3 1198811198732) (119881119873

3 1198811198733) (119881119873

3 1198811198734)

1198811198734

(1198811198734 1198811198731) (119881119873

4 1198811198732) (119881119873

4 1198811198733) (119881119873

4 1198811198734)



1 1199051⟩ rarr ⟨119873


119899 119905119899⟩)


1rarr 119873


119899)




119899) where

tr(119896+1)



tr(119896+1)


119896




1rarr 119873


119873119896minus1


1








Subregion 1

Subregion 2

Subregion 3

Subregion 4

Node 1Node 4

Node 2Node 3

Virtual node 4


Virtual node 1

Virtual node 4


Virtual node 1

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Physical topology

Logical topology


Link 2

Link 5

Link 1

Path 1 Path 2

Link 3

Link 4

Path 3 Path 4


Link 1


Link 5

Path 1

Path 2

Path 3

Path 4

Path 1Link 1

Link 3

Link 4

Path 2


Path 3Link 2

Link 3

Link 4 Path 2

Link 3Link 2


Path 1 Path 3

Path 4

Link 4

Path 2

(b) Index






[[[[[[[[

[

11990911

11990912


1199091119908

11990921

11990922


1199092119908


119909119879minus11

119909119879minus12


119909119879minus1119908

1199091198791

1199091198792


119909119879119908

]]]]]]]]

]

(1)








119885119871 119884119898

(119898 = 1 119871) (2)


119871 119884119898(119898 = 1 119871) in every scale is





119871 119898(119898 = 1 119871) at all scales



SPE119894=

119882

sum

119895=1

(119909119894119895minus 119909119894119895)2

(3)






1205852

120572= 1206011

[[

[

119888120572radic21206012ℎ2

0

1206011

+ 1 +1206012ℎ0(ℎ0minus 1)

1206012

1

]]

]

1ℎ0

(4)


1120601331206012

2 120601119894= sum119882

119895=119903+1120582119894

119895 119894 = 1 2 3 120582

119895is


120572is the 1 minus 120572





Transform

Physical topology

Logical topology

Taxi GPSdata

Traffic flowdata


transformPCA




Judge

Anomalousposition



119905

Thus

119876119905+1= 120573119876119905+ (1 minus 120573)119876

119905 (5)




120595 = 120583119904+ 119871 times 120590

119904radic

120573

(2 minus 120573) 119879 (6)
























119889119862119903 where 119862





GPS data



















1 432012 5927 6297 83172 632012 6418 6452 75863 732012 5344 7020 8849

32 1152012 6374 7270 8237

74 3182012 4728 7737 7888Average 119877 value 5633 6329 7462

050

100150200250300350400450500

Traffi

c flow


t

700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

20

40

60

80

100

120

140

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

50

100

150

200

250

300

350

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

10

20

30

40

50

60

70

80

()


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

845

ndash900

900

ndash915

915

ndash930











0

2000

4000

6000

8000

10000

12000

14000

16000


Proc

essin

g tim

e (m

s)



4 Conclusions












Acknowledgments



References











































1 Introduction










2 Literature Review









Organization


[35ndash43]

Project base


[36 37 40 41 43]

Systemtechnology


[35ndash43]

Consultants


[35ndash37 39]

Externalenvironment


[38 42]



3 Methodology








Organization





Project base





Systemtechnology






Consultant





Externalenvironment









Normalized Z(X)




each factor







DEMATEL

ANP





Z =

[[[[[[[

[

1199111111991112sdot sdot sdot 119911

1119899

1199112111991122sdot sdot sdot 119911

2119899

11991111989911199111198992sdot sdot sdot 119911

119899119899

]]]]]]]

]

(1)


120582 =

1

max1le119894le119899sum119899

119895=1119885119894119895

(119894 119895 = 1 2 119899)

X = 120582 sdot Z(2)


lim119870rarrinfin

X119870 = 0


(3)



Z =119860

119861

119862

119863

((

(

119860

0

3

3

3

119861

2

0

1

2

119862

2

2

0

2

119863

2

1

2

0

))

)

(4)


X =119860

119861

119862

119863

((

(

119860

0000

0337

0326

0337

119861

0233

0000

0116

0198

119862

0279

0198

0000

0198

119863

0233

0116

0244

0000

))

)

T =

119860

119861

119862

119863

(

119860

0628

0817

0839

0876

119861

0580

0356

0483

0559

119862

0691

0593

0449

0637

119863

0615

0493

0605

0424

)

119889

2513

2259

2377

2497

119903 3159 1979 2370 2137

(5)





Table 3


119860 2513 3159 5673 1 minus06462119861 2259 1979 4238 4 02796119862 2377 2370 4746 2 00068119863 2496 2137 4633 3 03598

Table 4

119860 119861 119862 119863

119860 0199 0293 0291 0288119861 0259 0180 0250 0231119862 0266 0244 0190 0283119863 0277 0283 0269 0199





Table 5


Overallrankings


A B C DA 0628 0580 0691 0615B 0817 0256 0593 0493C 0839 0483 0449 0605D 0876 0559 0637 0424

T =

Figure 2

DA

C

Figure 3

4 Empirical Study



























43 Result Analysis






Aspects 119860 119861 119862 119863

119860 00000 20000 24000 20000119861 29000 00000 17000 10000119862 28000 10000 00000 21000119863 29000 17000 17000 00000


Aspects 119860 119861 119862 119863

119860 00000 02326 02791 02326119861 03372 00000 01977 01163119862 03256 01163 00000 02442119863 03372 01977 01977 00000


Aspects 119860 119861 119862 119863 119889

119860 06278 05803 06905 06146 25132119861 08166 03563 05933 04925 22587119862 08389 04832 04492 06052 23765119863 08761 05593 06366 04242 24963119903 31593 19791 23697 21365


Aspects 119889 119903 119889 + 119903 119889 minus 119903

119860 25132 31593 56725 minus06462119861 22587 19791 42378 02796119862 23765 23697 47461 00068119863 24963 21365 46328 03598


Aspects 119860 119861 119862 119863

119860 01987 02932 02914 02877119861 02585 01800 02504 02305119862 02655 02442 01896 02832119863 02773 02826 02686 01986


Aspects 119860 119861 119862 119863

119860 02662 02662 02662 02662119861 02312 02312 02312 02312119862 02464 02464 02464 02464119863 02562 02562 02562 02562




Overallrankings



Organizations(A)


(D)System

technology (C)

Project base (B)









Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 40000 40000 40000 24000 20000 28000 40000 20000 40000 30000 400001198602 30000 00000 20000 18000 22000 20000 30000 00000 00000 00000 30000 200001198603 39000 20000 00000 30000 19000 21000 24000 25000 25000 36000 20000 220001198611 16000 27000 30000 00000 19000 30000 23000 20000 10000 17000 40000 290001198612 10000 16000 10000 10000 00000 30000 24000 10000 20000 24000 26000 180001198613 01000 15000 12000 02000 00000 00000 21000 00000 01000 04000 10000 140001198621 20000 18000 20000 14000 16000 10000 00000 30000 00000 00000 10000 300001198622 10000 10000 25000 14000 18000 19000 27000 00000 20000 25000 15000 140001198623 25000 20000 29000 20000 19000 20000 26000 30000 00000 29000 10000 200001198631 30000 30000 30000 08000 23000 30000 24000 00000 00000 00000 40000 300001198632 29000 20000 00000 06000 16000 26000 21000 09000 00000 31000 00000 130001198633 18000 13000 14000 02000 09000 03000 10000 00000 00000 00000 18000 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 01105 01105 01105 00663 00552 00773 01105 00552 01105 00829 011051198602 00829 00000 00552 00497 00608 00552 00829 00000 00000 00000 00829 005521198603 01077 00552 00000 00829 00525 00580 00663 00691 00691 00994 00552 006081198611 00442 00746 00829 00000 00525 00829 00635 00552 00276 00470 01105 008011198612 00276 00442 00276 00276 00000 00829 00663 00276 00552 00663 00718 004971198613 00028 00414 00331 00055 00000 00000 00580 00000 00028 00110 00276 003871198621 00552 00497 00552 00387 00442 00276 00000 00829 00000 00000 00276 008291198622 00276 00276 00691 00387 00497 00525 00746 00000 00552 00691 00414 003871198623 00691 00552 00801 00552 00525 00552 00718 00829 00000 00801 00276 005521198631 00829 00829 00829 00221 00635 00829 00663 00000 00000 00000 01105 008291198632 00801 00552 00000 00166 00442 00718 00580 00249 00000 00856 00000 003591198633 00497 00359 00387 00055 00249 00083 00276 00000 00000 00000 00497 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633 119889

1198601 01250 02233 02211 01894 01618 01718 02066 01854 01023 02070 02120 02347 224041198602 01424 00664 01129 00954 01090 01150 01484 00500 00274 00582 01475 01249 119751198603 01991 01544 01007 01508 01311 01526 01722 01371 01064 01808 01621 01682 181551198611 01294 01542 01563 00593 01173 01606 01537 01094 00602 01181 01938 01663 157861198612 00915 01064 00878 00699 00504 01407 01334 00697 00753 01158 01356 01170 119361198613 00316 00647 00553 00240 00212 00230 00828 00183 00112 00296 00533 00655 048041198621 01085 01029 01082 00795 00883 00807 00629 01188 00273 00512 00885 01398 105671198622 00962 00947 01311 00855 01019 01164 01447 00487 00806 01242 01120 01116 124771198623 01521 01393 01621 01165 01205 01368 01635 01403 00376 01511 01215 01482 158951198631 01614 01602 01518 00802 01243 01561 01513 00561 00320 00695 01910 01665 150021198632 01319 01132 00593 00575 00890 01249 01196 00625 00217 01277 00654 01007 107341198633 00816 00679 00671 00315 00508 00399 00624 00252 00143 00309 00824 00359 05899119903 14507 14476 14136 10395 11656 14185 16015 10217 05964 12641 15651 15790




3

Impo

rtan

ce ra

nkin

g

Noncritical

Critical1

7

8

12


















Criteria 119889 119903 119889 + 119903 119889 minus 119903




5 Conclusions









1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00862 01542 01564 01822 01388 01211 01290 01815 01715 01637 01355 014861198602 00982 00459 00799 00917 00935 00810 00927 00490 00459 00461 00943 007911198603 01372 01066 00712 01451 01125 01076 01075 01342 01784 01430 01036 010651198611 00892 01065 01105 00570 01007 01132 00960 01071 01009 00934 01238 010531198612 00631 00735 00621 00673 00432 00992 00833 00682 01263 00916 00866 007411198613 00218 00447 00391 00230 00182 00162 00517 00179 00188 00234 00341 004151198621 00748 00711 00765 00765 00757 00569 00393 01163 00458 00405 00566 008851198622 00663 00654 00927 00822 00874 00821 00904 00477 01352 00983 00716 007071198623 01048 00963 01147 01121 01034 00965 01021 01374 00630 01195 00776 009381198631 01112 01106 01074 00771 01066 01101 00945 00549 00537 00549 01220 010541198632 00909 00782 00420 00554 00764 00880 00747 00612 00364 01011 00418 006381198633 00562 00469 00474 00303 00436 00281 00390 00247 00240 00245 00527 00227


1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 014691198602 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 007491198603 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 012381198611 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 009801198612 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 007661198613 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 002851198621 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 006871198622 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 008381198623 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 010311198631 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 009061198632 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 006661198633 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386




























maintenance (C2)






Acknowledgments


References





















































1 Introduction













2 Literature Review












Potholeimages


Vehicle stationary


optics

Center





Center

RSE

company

OBU


Pothole information

Potholeinformation


or WAVE



severity)

or

















Input image

Binarization by HST

Segmented images



Candidaterefinement




Candidate region




119862V

=

1 if 119878 (1198721015840119888) gt 119879119904 Com (1198721015840


0 otherwise

(1)







com (1198721015840119888) =1198972

4120587119860 (2)



1198621015840

V =










OHI (ℎ119888 ℎ119887) =

119899

sum

119894=0

min (oh119894119888 oh119894119887) (4)









119901

=


1198881015840 ℎ119887) gt 119879119900 OHI (ℎ1015840

1198881015840 ℎ1015840



(5)




1198881015840 ℎ1015840





proposed method





ℎ(threshold for





119879ℎ


image119879std1015840 10

119879119904 512 119879ave 00119879com 005 119879

119900087

119879std 8 1198791199001015840 085





















(a) Asphalt images

(b) Concrete images







6 Conclusions





(a) Original images

(b) Close-up images



(a) Bright images

(b) Dark images








Acknowledgment


References






































Editorial Board










Contents















1 Introduction


















References
















1 Introduction













1198881 1198882 119888


1 1198882 119888

119899are


1 V2 V

119896 in



Custo

mer

-veh

icle

assig

nmen

t

Opt

imiz

ed as

signm

ent

CV

c1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

CV

c1c2

cn

12

k

middot



Fitness = min119899

sum

119894=0

119899

sum

119895=0

119896

sum

V=1119889119894119895119883

V119894119895+ 1198891198990119883

V1198990

119894 = 119895 (1)

119899

sum

119894=0

119899

sum

119895=0

119883

V119894119895119903119894le 119876V 119894 = 119895 V isin 119881 (2)

119883

V119894119895

=


0 otherwise

(3)


119894119895is the









1 1198882 119888

119899 to










119894= 119883

1198941 119883

119894119871





119894119871 whereas


119871


119881

new119894119895


times (119866119895minus 119883119894119895)

119883

new119894119895= 119883119894119895+ 119881

new119894119895

(4)



1

Depot

310

8

2

95

7

6

4

Opt

imiz

ed sc

hedu

le

Opt

imiz

ed as

signm

ent

1

Depot

72

8

10

95

3

6

4

7

Depot

310

8

5

92

1

6

4

CV

c1c2

cn

12

k

middot




119883

119881

119894119883

119881

1198941119883

119881


119881

119894119899119883

119881

119894119899+1119883

119881

119894119899+2sdot sdot sdot 119883

119881

119894119899+119896minus1











119894= 119883

119881

1198941 119883

119881

119894119899 119883

119881

119894119899+119896minus1


1Cus2 Cus

119899Veh1Veh2 Veh

119896minus1) For

















119863

1 119883

119863

2) is shown



Key2 13 08 24 19 02 12 21

02 08 12 13 19 2 21 24Key


Vehicle assignment

Visit order

Veh 1

Veh1

Veh1 Veh2

Veh2

Cus1

Cus1

Cus1

Veh 2

Cus2

Cus2

Cus2

Veh 3

Cus3

Cus3

Cus3

Cus4

Cus4

Cus4

Cus5

Cus5

Cus5

Cus6

Cus6

Cus6

XiV

XiV



X119863 119883

119863

1119883

119863

2






119862 119910119862)) can be yielded by

119909119862=

sum

119896

119894=1119909

V119894+ 119909119874

119896 + 1

119910119862=

sum

119896

119894=1119910

V119894+ 119910119874

119896 + 1

(5)

F

O

DE

G

HA

I

C

J

B

K

Route-1

Route-2

Route-3


F

O

DE

G

HA

I

C

J

B

K

dE

dF

dD

dC

dB

dA

CC


In (5) 119909119862and 119910





119874are the




F

O

DE

B

C

JK

G

H

I

A

(a) 1198891 = 119874119861 + 119861119866minus 119874119866

F

O

DE

B

C

JK

G

H

I

A

(b) 1198892 = 119866119861 + 119861119867minus 119866119867

F

O

DE

C

J

A

K

G

H

IB

(c) 1198893 = 119867119861 + 119861119868 minus 119867119868

F

O

DE

B

C

J

A

K

G

H

I

(d) 1198894 = 119868119861 + 119861119874minus 119868119874








(3) Outer layer PSO



layer PSO





Start




End

Yes Yes

NoNo

YesNo


Initialize VVX

V

Update VVX

V

Initialize VDX

D

Update VDX

D

search(XV)

Fitness(X ) lt

Fitness(XV)

Update(SA)

Fitness( )

Updateand

Pass XD

to inner layer PSO

Fitness(XD) =

Fitness( )= XLSV

GVbest

XVnew

PVbest

XVnew X

Vnew

Updateand

GVbest

PVbest

GVbest

LSV

XVLS = local




119899









A-n54-k9 53 + 53 53 + 8

A-n64-k8 63 + 63 63 + 7







1= 2 and 119888

2= 1 and the


1= 1198882= 2 and 100



BKStimes 100 (6)


Avg Dev () =sum

119899

119894=1DEV119894

119899

(7)




1) RBI (LS

2)




0102030405060708090

Aver

age d

evia

tion

()

A-n3

2-k5

A-n3

3-k5

A-n3

6-k5

A-n4

5-k6

A-n4

5-k7

A-n5

5-k9

A-n6

0-k9

A-n6

2-k8

A-n6

4-k9

Aver

age

wo LSLS1

LS2LS3




Imp() =Fitness


Fitness119908119900

times 100 (8)





Fitn

ess

950

900

850

800

750

700

Iterations

0

50

100

150

200

250

300

350

400

450

500

550

600

650

700

750

800

850

900

950

1000

(a)

Fitn

ess

830

810

790

770

750

730

710

690

670

650

Iterations

0 5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

(b)


(a) (b)






(a) (b)



5 Conclusions






Acknowledgment


References




































1 Introduction



























HMM correction

Vehicle V1

Vehicle V2

Vehicle Vn



based on HMM

Entireroutes

Routerecommendation


Vehicle Vi

A set ofOutput

Input

RS(Tk Roadi)

RouteT119896



upcomingroutes


Rm1Rm2

Rm3

Rm4

Rm5

Rm1x

Rm2y

Rm3x Rm4x

Rm5y

Y

X0


Ri

Rj

Rk

Rix

Rjy

Rkx

Rky

Y

X0



119860119899

997888rarr 119861 = 1198771119909 (1198771119910)


(1)





= (1 3) (1 4) (3 4) (3 1)

(2)






Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd


Sichuan RdJinmao St

Longpan Rd

Jianning Rd

0 1 2 3 4 5 6 7 80

1

2

3

4

5

6

Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd

Central Rd

Xufu Alley

Sichuan Rd

Jinmao St

Longpan Rd

Jianning Rd

X

Y












1199051 lt (1 3) (1 4) (3 4) (3 1) gt

1199052 lt (3 1) (3 4) (1 4) (1 3) gt

1199053 lt (0 3) (1 3) (1 5) (4 5) gt

1199054 lt (0 3) (0 0) (0 4) (4 1) gt

1199055 lt (2 0) (2 1) (3 1) (3 2) (4 2) gt

1199051 lt (1 3) (1 4) (3 4) (3 1) gt

(3)














997888997888997888997888rarr119860119895119861119895 = (119909119895119899minus1199091198951 119910119895119899minus



cos ⟨ 119886119894 119886119895⟩ =

119886119894 sdot 119886119895

1003816100381610038161003816 1198861198941003816100381610038161003816 sdot10038161003816100381610038161003816119886119895

10038161003816100381610038161003816

= ((119909119894119899 minus 1199091198941) sdot (119910119894119899 minus 1199101198941)

+ (119909119895119899 minus 1199091198951) sdot (119910119895119899 minus 1199101198951))


2

sdotradic(119909119895119899 minus 1199091198951)2

+ (119910119895119899 minus 1199101198951)2

)

minus1

(4)


1





120587119894 =

Count (119904119863119894

) + 120582Count (119904New119863119894

)

sum119899

119895=1[Count (119904119863

119895

) + 120582Count (119904New119863119895

)]

(5)




1198961= 1199001198962)






119894

)

and Count(119904New119863119894



119875 (119904119894 | 119904119894minus1)

=

Count (119904119863119894minus1

119904119863119894

) + 120582Count (119904New119863119894minus1

119904New119863119894

) + 1

Count (119904119863119894minus1

) + 120582Count (119904New119863119894minus1

) + 119898

(6)


119904119863119894


119904New119863119894

)



119875 (V119895 | 119904119894)

=

Count (119904119863119894minus1

V119863119894

) + 120582Count (119904New119863119894minus1

VNew119863119894

) + 1

Count (119904119863119894

) + 120582Count (119904New119863119894

) + 119899

(7)


V119863119894


VNew119863119894

)






sequence

Viterbialgorithm










119901119894(1199091) 119901119894(119909

2) 119901119894(119909



1) 119901119894(119910

2) 119901119894(119910

119899) (1198632 notin 119881) and the











(b) Suppose that 1198632 = 119901119894(1199101) 119901119894(119910

2) 119901119894119896 then


(c) Suppose that 1198632 = 1199011198941 119901119894(1199102) 119901119894(119910

119899) then



1) 119901119894(119910

2) 119901119894(119910

119899) and 1199101 = 1 119910119899


1)

119901119894(1199102) 119901119894(119910





















119875119894 =sum119899

119896=1119863(119877119896 119862119877)

sum119899

119905=1Dist 1003816100381610038161003816119877119905

1003816100381610038161003816

times 100 (8)



1198751 =Repeat (1198771 1198771) + Repeat (1198773 1198771)

Dist 100381610038161003816100381611987711003816100381610038161003816 + Dist 10038161003816100381610038161198773

1003816100381610038161003816

times 100

=6 + 1

6 + 7times 100 = 5385

(9)












(a) (b)


100908070605040302010

01009080706050403020100

Trip completed ()

Cor

rect

pre

dict

ion

()


Cor

rect

pre

dict

ion

()

100908070605040302010

01009080706050403020100

Trip completed ()


Cor

rect

pre

dict

ion

()


100908070605040302010

0


120582 = 05

120582 = 02


Cor

rect

pre

dict

ion

()

100908070605040302010

0



120582 = 05

120582 = 02







90

80

70

60

50

40

30

20

10

0Other time periods

Cor

rect

pre

dict

ion

()

Time of day







7 Conclusion






Acknowledgments


References

























1 Introduction



















1198811198732

1198811198733

1198811198734

1198811198731

(1198811198731 1198811198731) (119881119873

1 1198811198732) (119881119873

1 1198811198733) (119881119873

1 1198811198734)

1198811198732

(1198811198732 1198811198731) (119881119873

2 1198811198732) (119881119873

2 1198811198733) (119881119873

2 1198811198734)

1198811198733

(1198811198733 1198811198731) (119881119873

3 1198811198732) (119881119873

3 1198811198733) (119881119873

3 1198811198734)

1198811198734

(1198811198734 1198811198731) (119881119873

4 1198811198732) (119881119873

4 1198811198733) (119881119873

4 1198811198734)



1 1199051⟩ rarr ⟨119873


119899 119905119899⟩)


1rarr 119873


119899)




119899) where

tr(119896+1)



tr(119896+1)


119896




1rarr 119873


119873119896minus1


1








Subregion 1

Subregion 2

Subregion 3

Subregion 4

Node 1Node 4

Node 2Node 3

Virtual node 4


Virtual node 1

Virtual node 4


Virtual node 1

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Physical topology

Logical topology


Link 2

Link 5

Link 1

Path 1 Path 2

Link 3

Link 4

Path 3 Path 4


Link 1


Link 5

Path 1

Path 2

Path 3

Path 4

Path 1Link 1

Link 3

Link 4

Path 2


Path 3Link 2

Link 3

Link 4 Path 2

Link 3Link 2


Path 1 Path 3

Path 4

Link 4

Path 2

(b) Index






[[[[[[[[

[

11990911

11990912


1199091119908

11990921

11990922


1199092119908


119909119879minus11

119909119879minus12


119909119879minus1119908

1199091198791

1199091198792


119909119879119908

]]]]]]]]

]

(1)








119885119871 119884119898

(119898 = 1 119871) (2)


119871 119884119898(119898 = 1 119871) in every scale is





119871 119898(119898 = 1 119871) at all scales



SPE119894=

119882

sum

119895=1

(119909119894119895minus 119909119894119895)2

(3)






1205852

120572= 1206011

[[

[

119888120572radic21206012ℎ2

0

1206011

+ 1 +1206012ℎ0(ℎ0minus 1)

1206012

1

]]

]

1ℎ0

(4)


1120601331206012

2 120601119894= sum119882

119895=119903+1120582119894

119895 119894 = 1 2 3 120582

119895is


120572is the 1 minus 120572





Transform

Physical topology

Logical topology

Taxi GPSdata

Traffic flowdata


transformPCA




Judge

Anomalousposition



119905

Thus

119876119905+1= 120573119876119905+ (1 minus 120573)119876

119905 (5)




120595 = 120583119904+ 119871 times 120590

119904radic

120573

(2 minus 120573) 119879 (6)
























119889119862119903 where 119862





GPS data



















1 432012 5927 6297 83172 632012 6418 6452 75863 732012 5344 7020 8849

32 1152012 6374 7270 8237

74 3182012 4728 7737 7888Average 119877 value 5633 6329 7462

050

100150200250300350400450500

Traffi

c flow


t

700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

20

40

60

80

100

120

140

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

50

100

150

200

250

300

350

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

10

20

30

40

50

60

70

80

()


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

845

ndash900

900

ndash915

915

ndash930











0

2000

4000

6000

8000

10000

12000

14000

16000


Proc

essin

g tim

e (m

s)



4 Conclusions












Acknowledgments



References











































1 Introduction










2 Literature Review









Organization


[35ndash43]

Project base


[36 37 40 41 43]

Systemtechnology


[35ndash43]

Consultants


[35ndash37 39]

Externalenvironment


[38 42]



3 Methodology








Organization





Project base





Systemtechnology






Consultant





Externalenvironment









Normalized Z(X)




each factor







DEMATEL

ANP





Z =

[[[[[[[

[

1199111111991112sdot sdot sdot 119911

1119899

1199112111991122sdot sdot sdot 119911

2119899

11991111989911199111198992sdot sdot sdot 119911

119899119899

]]]]]]]

]

(1)


120582 =

1

max1le119894le119899sum119899

119895=1119885119894119895

(119894 119895 = 1 2 119899)

X = 120582 sdot Z(2)


lim119870rarrinfin

X119870 = 0


(3)



Z =119860

119861

119862

119863

((

(

119860

0

3

3

3

119861

2

0

1

2

119862

2

2

0

2

119863

2

1

2

0

))

)

(4)


X =119860

119861

119862

119863

((

(

119860

0000

0337

0326

0337

119861

0233

0000

0116

0198

119862

0279

0198

0000

0198

119863

0233

0116

0244

0000

))

)

T =

119860

119861

119862

119863

(

119860

0628

0817

0839

0876

119861

0580

0356

0483

0559

119862

0691

0593

0449

0637

119863

0615

0493

0605

0424

)

119889

2513

2259

2377

2497

119903 3159 1979 2370 2137

(5)





Table 3


119860 2513 3159 5673 1 minus06462119861 2259 1979 4238 4 02796119862 2377 2370 4746 2 00068119863 2496 2137 4633 3 03598

Table 4

119860 119861 119862 119863

119860 0199 0293 0291 0288119861 0259 0180 0250 0231119862 0266 0244 0190 0283119863 0277 0283 0269 0199





Table 5


Overallrankings


A B C DA 0628 0580 0691 0615B 0817 0256 0593 0493C 0839 0483 0449 0605D 0876 0559 0637 0424

T =

Figure 2

DA

C

Figure 3

4 Empirical Study



























43 Result Analysis






Aspects 119860 119861 119862 119863

119860 00000 20000 24000 20000119861 29000 00000 17000 10000119862 28000 10000 00000 21000119863 29000 17000 17000 00000


Aspects 119860 119861 119862 119863

119860 00000 02326 02791 02326119861 03372 00000 01977 01163119862 03256 01163 00000 02442119863 03372 01977 01977 00000


Aspects 119860 119861 119862 119863 119889

119860 06278 05803 06905 06146 25132119861 08166 03563 05933 04925 22587119862 08389 04832 04492 06052 23765119863 08761 05593 06366 04242 24963119903 31593 19791 23697 21365


Aspects 119889 119903 119889 + 119903 119889 minus 119903

119860 25132 31593 56725 minus06462119861 22587 19791 42378 02796119862 23765 23697 47461 00068119863 24963 21365 46328 03598


Aspects 119860 119861 119862 119863

119860 01987 02932 02914 02877119861 02585 01800 02504 02305119862 02655 02442 01896 02832119863 02773 02826 02686 01986


Aspects 119860 119861 119862 119863

119860 02662 02662 02662 02662119861 02312 02312 02312 02312119862 02464 02464 02464 02464119863 02562 02562 02562 02562




Overallrankings



Organizations(A)


(D)System

technology (C)

Project base (B)









Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 40000 40000 40000 24000 20000 28000 40000 20000 40000 30000 400001198602 30000 00000 20000 18000 22000 20000 30000 00000 00000 00000 30000 200001198603 39000 20000 00000 30000 19000 21000 24000 25000 25000 36000 20000 220001198611 16000 27000 30000 00000 19000 30000 23000 20000 10000 17000 40000 290001198612 10000 16000 10000 10000 00000 30000 24000 10000 20000 24000 26000 180001198613 01000 15000 12000 02000 00000 00000 21000 00000 01000 04000 10000 140001198621 20000 18000 20000 14000 16000 10000 00000 30000 00000 00000 10000 300001198622 10000 10000 25000 14000 18000 19000 27000 00000 20000 25000 15000 140001198623 25000 20000 29000 20000 19000 20000 26000 30000 00000 29000 10000 200001198631 30000 30000 30000 08000 23000 30000 24000 00000 00000 00000 40000 300001198632 29000 20000 00000 06000 16000 26000 21000 09000 00000 31000 00000 130001198633 18000 13000 14000 02000 09000 03000 10000 00000 00000 00000 18000 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 01105 01105 01105 00663 00552 00773 01105 00552 01105 00829 011051198602 00829 00000 00552 00497 00608 00552 00829 00000 00000 00000 00829 005521198603 01077 00552 00000 00829 00525 00580 00663 00691 00691 00994 00552 006081198611 00442 00746 00829 00000 00525 00829 00635 00552 00276 00470 01105 008011198612 00276 00442 00276 00276 00000 00829 00663 00276 00552 00663 00718 004971198613 00028 00414 00331 00055 00000 00000 00580 00000 00028 00110 00276 003871198621 00552 00497 00552 00387 00442 00276 00000 00829 00000 00000 00276 008291198622 00276 00276 00691 00387 00497 00525 00746 00000 00552 00691 00414 003871198623 00691 00552 00801 00552 00525 00552 00718 00829 00000 00801 00276 005521198631 00829 00829 00829 00221 00635 00829 00663 00000 00000 00000 01105 008291198632 00801 00552 00000 00166 00442 00718 00580 00249 00000 00856 00000 003591198633 00497 00359 00387 00055 00249 00083 00276 00000 00000 00000 00497 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633 119889

1198601 01250 02233 02211 01894 01618 01718 02066 01854 01023 02070 02120 02347 224041198602 01424 00664 01129 00954 01090 01150 01484 00500 00274 00582 01475 01249 119751198603 01991 01544 01007 01508 01311 01526 01722 01371 01064 01808 01621 01682 181551198611 01294 01542 01563 00593 01173 01606 01537 01094 00602 01181 01938 01663 157861198612 00915 01064 00878 00699 00504 01407 01334 00697 00753 01158 01356 01170 119361198613 00316 00647 00553 00240 00212 00230 00828 00183 00112 00296 00533 00655 048041198621 01085 01029 01082 00795 00883 00807 00629 01188 00273 00512 00885 01398 105671198622 00962 00947 01311 00855 01019 01164 01447 00487 00806 01242 01120 01116 124771198623 01521 01393 01621 01165 01205 01368 01635 01403 00376 01511 01215 01482 158951198631 01614 01602 01518 00802 01243 01561 01513 00561 00320 00695 01910 01665 150021198632 01319 01132 00593 00575 00890 01249 01196 00625 00217 01277 00654 01007 107341198633 00816 00679 00671 00315 00508 00399 00624 00252 00143 00309 00824 00359 05899119903 14507 14476 14136 10395 11656 14185 16015 10217 05964 12641 15651 15790




3

Impo

rtan

ce ra

nkin

g

Noncritical

Critical1

7

8

12


















Criteria 119889 119903 119889 + 119903 119889 minus 119903




5 Conclusions









1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00862 01542 01564 01822 01388 01211 01290 01815 01715 01637 01355 014861198602 00982 00459 00799 00917 00935 00810 00927 00490 00459 00461 00943 007911198603 01372 01066 00712 01451 01125 01076 01075 01342 01784 01430 01036 010651198611 00892 01065 01105 00570 01007 01132 00960 01071 01009 00934 01238 010531198612 00631 00735 00621 00673 00432 00992 00833 00682 01263 00916 00866 007411198613 00218 00447 00391 00230 00182 00162 00517 00179 00188 00234 00341 004151198621 00748 00711 00765 00765 00757 00569 00393 01163 00458 00405 00566 008851198622 00663 00654 00927 00822 00874 00821 00904 00477 01352 00983 00716 007071198623 01048 00963 01147 01121 01034 00965 01021 01374 00630 01195 00776 009381198631 01112 01106 01074 00771 01066 01101 00945 00549 00537 00549 01220 010541198632 00909 00782 00420 00554 00764 00880 00747 00612 00364 01011 00418 006381198633 00562 00469 00474 00303 00436 00281 00390 00247 00240 00245 00527 00227


1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 014691198602 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 007491198603 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 012381198611 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 009801198612 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 007661198613 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 002851198621 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 006871198622 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 008381198623 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 010311198631 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 009061198632 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 006661198633 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386




























maintenance (C2)






Acknowledgments


References





















































1 Introduction













2 Literature Review












Potholeimages


Vehicle stationary


optics

Center





Center

RSE

company

OBU


Pothole information

Potholeinformation


or WAVE



severity)

or

















Input image

Binarization by HST

Segmented images



Candidaterefinement




Candidate region




119862V

=

1 if 119878 (1198721015840119888) gt 119879119904 Com (1198721015840


0 otherwise

(1)







com (1198721015840119888) =1198972

4120587119860 (2)



1198621015840

V =










OHI (ℎ119888 ℎ119887) =

119899

sum

119894=0

min (oh119894119888 oh119894119887) (4)









119901

=


1198881015840 ℎ119887) gt 119879119900 OHI (ℎ1015840

1198881015840 ℎ1015840



(5)




1198881015840 ℎ1015840





proposed method





ℎ(threshold for





119879ℎ


image119879std1015840 10

119879119904 512 119879ave 00119879com 005 119879

119900087

119879std 8 1198791199001015840 085





















(a) Asphalt images

(b) Concrete images







6 Conclusions





(a) Original images

(b) Close-up images



(a) Bright images

(b) Dark images








Acknowledgment


References

































Editorial Board










Contents















1 Introduction


















References
















1 Introduction













1198881 1198882 119888


1 1198882 119888

119899are


1 V2 V

119896 in



Custo

mer

-veh

icle

assig

nmen

t

Opt

imiz

ed as

signm

ent

CV

c1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

CV

c1c2

cn

12

k

middot



Fitness = min119899

sum

119894=0

119899

sum

119895=0

119896

sum

V=1119889119894119895119883

V119894119895+ 1198891198990119883

V1198990

119894 = 119895 (1)

119899

sum

119894=0

119899

sum

119895=0

119883

V119894119895119903119894le 119876V 119894 = 119895 V isin 119881 (2)

119883

V119894119895

=


0 otherwise

(3)


119894119895is the









1 1198882 119888

119899 to










119894= 119883

1198941 119883

119894119871





119894119871 whereas


119871


119881

new119894119895


times (119866119895minus 119883119894119895)

119883

new119894119895= 119883119894119895+ 119881

new119894119895

(4)



1

Depot

310

8

2

95

7

6

4

Opt

imiz

ed sc

hedu

le

Opt

imiz

ed as

signm

ent

1

Depot

72

8

10

95

3

6

4

7

Depot

310

8

5

92

1

6

4

CV

c1c2

cn

12

k

middot




119883

119881

119894119883

119881

1198941119883

119881


119881

119894119899119883

119881

119894119899+1119883

119881

119894119899+2sdot sdot sdot 119883

119881

119894119899+119896minus1











119894= 119883

119881

1198941 119883

119881

119894119899 119883

119881

119894119899+119896minus1


1Cus2 Cus

119899Veh1Veh2 Veh

119896minus1) For

















119863

1 119883

119863

2) is shown



Key2 13 08 24 19 02 12 21

02 08 12 13 19 2 21 24Key


Vehicle assignment

Visit order

Veh 1

Veh1

Veh1 Veh2

Veh2

Cus1

Cus1

Cus1

Veh 2

Cus2

Cus2

Cus2

Veh 3

Cus3

Cus3

Cus3

Cus4

Cus4

Cus4

Cus5

Cus5

Cus5

Cus6

Cus6

Cus6

XiV

XiV



X119863 119883

119863

1119883

119863

2






119862 119910119862)) can be yielded by

119909119862=

sum

119896

119894=1119909

V119894+ 119909119874

119896 + 1

119910119862=

sum

119896

119894=1119910

V119894+ 119910119874

119896 + 1

(5)

F

O

DE

G

HA

I

C

J

B

K

Route-1

Route-2

Route-3


F

O

DE

G

HA

I

C

J

B

K

dE

dF

dD

dC

dB

dA

CC


In (5) 119909119862and 119910





119874are the




F

O

DE

B

C

JK

G

H

I

A

(a) 1198891 = 119874119861 + 119861119866minus 119874119866

F

O

DE

B

C

JK

G

H

I

A

(b) 1198892 = 119866119861 + 119861119867minus 119866119867

F

O

DE

C

J

A

K

G

H

IB

(c) 1198893 = 119867119861 + 119861119868 minus 119867119868

F

O

DE

B

C

J

A

K

G

H

I

(d) 1198894 = 119868119861 + 119861119874minus 119868119874








(3) Outer layer PSO



layer PSO





Start




End

Yes Yes

NoNo

YesNo


Initialize VVX

V

Update VVX

V

Initialize VDX

D

Update VDX

D

search(XV)

Fitness(X ) lt

Fitness(XV)

Update(SA)

Fitness( )

Updateand

Pass XD

to inner layer PSO

Fitness(XD) =

Fitness( )= XLSV

GVbest

XVnew

PVbest

XVnew X

Vnew

Updateand

GVbest

PVbest

GVbest

LSV

XVLS = local




119899









A-n54-k9 53 + 53 53 + 8

A-n64-k8 63 + 63 63 + 7







1= 2 and 119888

2= 1 and the


1= 1198882= 2 and 100



BKStimes 100 (6)


Avg Dev () =sum

119899

119894=1DEV119894

119899

(7)




1) RBI (LS

2)




0102030405060708090

Aver

age d

evia

tion

()

A-n3

2-k5

A-n3

3-k5

A-n3

6-k5

A-n4

5-k6

A-n4

5-k7

A-n5

5-k9

A-n6

0-k9

A-n6

2-k8

A-n6

4-k9

Aver

age

wo LSLS1

LS2LS3




Imp() =Fitness


Fitness119908119900

times 100 (8)





Fitn

ess

950

900

850

800

750

700

Iterations

0

50

100

150

200

250

300

350

400

450

500

550

600

650

700

750

800

850

900

950

1000

(a)

Fitn

ess

830

810

790

770

750

730

710

690

670

650

Iterations

0 5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

(b)


(a) (b)






(a) (b)



5 Conclusions






Acknowledgment


References




































1 Introduction



























HMM correction

Vehicle V1

Vehicle V2

Vehicle Vn



based on HMM

Entireroutes

Routerecommendation


Vehicle Vi

A set ofOutput

Input

RS(Tk Roadi)

RouteT119896



upcomingroutes


Rm1Rm2

Rm3

Rm4

Rm5

Rm1x

Rm2y

Rm3x Rm4x

Rm5y

Y

X0


Ri

Rj

Rk

Rix

Rjy

Rkx

Rky

Y

X0



119860119899

997888rarr 119861 = 1198771119909 (1198771119910)


(1)





= (1 3) (1 4) (3 4) (3 1)

(2)






Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd


Sichuan RdJinmao St

Longpan Rd

Jianning Rd

0 1 2 3 4 5 6 7 80

1

2

3

4

5

6

Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd

Central Rd

Xufu Alley

Sichuan Rd

Jinmao St

Longpan Rd

Jianning Rd

X

Y












1199051 lt (1 3) (1 4) (3 4) (3 1) gt

1199052 lt (3 1) (3 4) (1 4) (1 3) gt

1199053 lt (0 3) (1 3) (1 5) (4 5) gt

1199054 lt (0 3) (0 0) (0 4) (4 1) gt

1199055 lt (2 0) (2 1) (3 1) (3 2) (4 2) gt

1199051 lt (1 3) (1 4) (3 4) (3 1) gt

(3)














997888997888997888997888rarr119860119895119861119895 = (119909119895119899minus1199091198951 119910119895119899minus



cos ⟨ 119886119894 119886119895⟩ =

119886119894 sdot 119886119895

1003816100381610038161003816 1198861198941003816100381610038161003816 sdot10038161003816100381610038161003816119886119895

10038161003816100381610038161003816

= ((119909119894119899 minus 1199091198941) sdot (119910119894119899 minus 1199101198941)

+ (119909119895119899 minus 1199091198951) sdot (119910119895119899 minus 1199101198951))


2

sdotradic(119909119895119899 minus 1199091198951)2

+ (119910119895119899 minus 1199101198951)2

)

minus1

(4)


1





120587119894 =

Count (119904119863119894

) + 120582Count (119904New119863119894

)

sum119899

119895=1[Count (119904119863

119895

) + 120582Count (119904New119863119895

)]

(5)




1198961= 1199001198962)






119894

)

and Count(119904New119863119894



119875 (119904119894 | 119904119894minus1)

=

Count (119904119863119894minus1

119904119863119894

) + 120582Count (119904New119863119894minus1

119904New119863119894

) + 1

Count (119904119863119894minus1

) + 120582Count (119904New119863119894minus1

) + 119898

(6)


119904119863119894


119904New119863119894

)



119875 (V119895 | 119904119894)

=

Count (119904119863119894minus1

V119863119894

) + 120582Count (119904New119863119894minus1

VNew119863119894

) + 1

Count (119904119863119894

) + 120582Count (119904New119863119894

) + 119899

(7)


V119863119894


VNew119863119894

)






sequence

Viterbialgorithm










119901119894(1199091) 119901119894(119909

2) 119901119894(119909



1) 119901119894(119910

2) 119901119894(119910

119899) (1198632 notin 119881) and the











(b) Suppose that 1198632 = 119901119894(1199101) 119901119894(119910

2) 119901119894119896 then


(c) Suppose that 1198632 = 1199011198941 119901119894(1199102) 119901119894(119910

119899) then



1) 119901119894(119910

2) 119901119894(119910

119899) and 1199101 = 1 119910119899


1)

119901119894(1199102) 119901119894(119910





















119875119894 =sum119899

119896=1119863(119877119896 119862119877)

sum119899

119905=1Dist 1003816100381610038161003816119877119905

1003816100381610038161003816

times 100 (8)



1198751 =Repeat (1198771 1198771) + Repeat (1198773 1198771)

Dist 100381610038161003816100381611987711003816100381610038161003816 + Dist 10038161003816100381610038161198773

1003816100381610038161003816

times 100

=6 + 1

6 + 7times 100 = 5385

(9)












(a) (b)


100908070605040302010

01009080706050403020100

Trip completed ()

Cor

rect

pre

dict

ion

()


Cor

rect

pre

dict

ion

()

100908070605040302010

01009080706050403020100

Trip completed ()


Cor

rect

pre

dict

ion

()


100908070605040302010

0


120582 = 05

120582 = 02


Cor

rect

pre

dict

ion

()

100908070605040302010

0



120582 = 05

120582 = 02







90

80

70

60

50

40

30

20

10

0Other time periods

Cor

rect

pre

dict

ion

()

Time of day







7 Conclusion






Acknowledgments


References

























1 Introduction



















1198811198732

1198811198733

1198811198734

1198811198731

(1198811198731 1198811198731) (119881119873

1 1198811198732) (119881119873

1 1198811198733) (119881119873

1 1198811198734)

1198811198732

(1198811198732 1198811198731) (119881119873

2 1198811198732) (119881119873

2 1198811198733) (119881119873

2 1198811198734)

1198811198733

(1198811198733 1198811198731) (119881119873

3 1198811198732) (119881119873

3 1198811198733) (119881119873

3 1198811198734)

1198811198734

(1198811198734 1198811198731) (119881119873

4 1198811198732) (119881119873

4 1198811198733) (119881119873

4 1198811198734)



1 1199051⟩ rarr ⟨119873


119899 119905119899⟩)


1rarr 119873


119899)




119899) where

tr(119896+1)



tr(119896+1)


119896




1rarr 119873


119873119896minus1


1








Subregion 1

Subregion 2

Subregion 3

Subregion 4

Node 1Node 4

Node 2Node 3

Virtual node 4


Virtual node 1

Virtual node 4


Virtual node 1

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Physical topology

Logical topology


Link 2

Link 5

Link 1

Path 1 Path 2

Link 3

Link 4

Path 3 Path 4


Link 1


Link 5

Path 1

Path 2

Path 3

Path 4

Path 1Link 1

Link 3

Link 4

Path 2


Path 3Link 2

Link 3

Link 4 Path 2

Link 3Link 2


Path 1 Path 3

Path 4

Link 4

Path 2

(b) Index






[[[[[[[[

[

11990911

11990912


1199091119908

11990921

11990922


1199092119908


119909119879minus11

119909119879minus12


119909119879minus1119908

1199091198791

1199091198792


119909119879119908

]]]]]]]]

]

(1)








119885119871 119884119898

(119898 = 1 119871) (2)


119871 119884119898(119898 = 1 119871) in every scale is





119871 119898(119898 = 1 119871) at all scales



SPE119894=

119882

sum

119895=1

(119909119894119895minus 119909119894119895)2

(3)






1205852

120572= 1206011

[[

[

119888120572radic21206012ℎ2

0

1206011

+ 1 +1206012ℎ0(ℎ0minus 1)

1206012

1

]]

]

1ℎ0

(4)


1120601331206012

2 120601119894= sum119882

119895=119903+1120582119894

119895 119894 = 1 2 3 120582

119895is


120572is the 1 minus 120572





Transform

Physical topology

Logical topology

Taxi GPSdata

Traffic flowdata


transformPCA




Judge

Anomalousposition



119905

Thus

119876119905+1= 120573119876119905+ (1 minus 120573)119876

119905 (5)




120595 = 120583119904+ 119871 times 120590

119904radic

120573

(2 minus 120573) 119879 (6)
























119889119862119903 where 119862





GPS data



















1 432012 5927 6297 83172 632012 6418 6452 75863 732012 5344 7020 8849

32 1152012 6374 7270 8237

74 3182012 4728 7737 7888Average 119877 value 5633 6329 7462

050

100150200250300350400450500

Traffi

c flow


t

700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

20

40

60

80

100

120

140

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

50

100

150

200

250

300

350

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

10

20

30

40

50

60

70

80

()


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

845

ndash900

900

ndash915

915

ndash930











0

2000

4000

6000

8000

10000

12000

14000

16000


Proc

essin

g tim

e (m

s)



4 Conclusions












Acknowledgments



References











































1 Introduction










2 Literature Review









Organization


[35ndash43]

Project base


[36 37 40 41 43]

Systemtechnology


[35ndash43]

Consultants


[35ndash37 39]

Externalenvironment


[38 42]



3 Methodology








Organization





Project base





Systemtechnology






Consultant





Externalenvironment









Normalized Z(X)




each factor







DEMATEL

ANP





Z =

[[[[[[[

[

1199111111991112sdot sdot sdot 119911

1119899

1199112111991122sdot sdot sdot 119911

2119899

11991111989911199111198992sdot sdot sdot 119911

119899119899

]]]]]]]

]

(1)


120582 =

1

max1le119894le119899sum119899

119895=1119885119894119895

(119894 119895 = 1 2 119899)

X = 120582 sdot Z(2)


lim119870rarrinfin

X119870 = 0


(3)



Z =119860

119861

119862

119863

((

(

119860

0

3

3

3

119861

2

0

1

2

119862

2

2

0

2

119863

2

1

2

0

))

)

(4)


X =119860

119861

119862

119863

((

(

119860

0000

0337

0326

0337

119861

0233

0000

0116

0198

119862

0279

0198

0000

0198

119863

0233

0116

0244

0000

))

)

T =

119860

119861

119862

119863

(

119860

0628

0817

0839

0876

119861

0580

0356

0483

0559

119862

0691

0593

0449

0637

119863

0615

0493

0605

0424

)

119889

2513

2259

2377

2497

119903 3159 1979 2370 2137

(5)





Table 3


119860 2513 3159 5673 1 minus06462119861 2259 1979 4238 4 02796119862 2377 2370 4746 2 00068119863 2496 2137 4633 3 03598

Table 4

119860 119861 119862 119863

119860 0199 0293 0291 0288119861 0259 0180 0250 0231119862 0266 0244 0190 0283119863 0277 0283 0269 0199





Table 5


Overallrankings


A B C DA 0628 0580 0691 0615B 0817 0256 0593 0493C 0839 0483 0449 0605D 0876 0559 0637 0424

T =

Figure 2

DA

C

Figure 3

4 Empirical Study



























43 Result Analysis






Aspects 119860 119861 119862 119863

119860 00000 20000 24000 20000119861 29000 00000 17000 10000119862 28000 10000 00000 21000119863 29000 17000 17000 00000


Aspects 119860 119861 119862 119863

119860 00000 02326 02791 02326119861 03372 00000 01977 01163119862 03256 01163 00000 02442119863 03372 01977 01977 00000


Aspects 119860 119861 119862 119863 119889

119860 06278 05803 06905 06146 25132119861 08166 03563 05933 04925 22587119862 08389 04832 04492 06052 23765119863 08761 05593 06366 04242 24963119903 31593 19791 23697 21365


Aspects 119889 119903 119889 + 119903 119889 minus 119903

119860 25132 31593 56725 minus06462119861 22587 19791 42378 02796119862 23765 23697 47461 00068119863 24963 21365 46328 03598


Aspects 119860 119861 119862 119863

119860 01987 02932 02914 02877119861 02585 01800 02504 02305119862 02655 02442 01896 02832119863 02773 02826 02686 01986


Aspects 119860 119861 119862 119863

119860 02662 02662 02662 02662119861 02312 02312 02312 02312119862 02464 02464 02464 02464119863 02562 02562 02562 02562




Overallrankings



Organizations(A)


(D)System

technology (C)

Project base (B)









Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 40000 40000 40000 24000 20000 28000 40000 20000 40000 30000 400001198602 30000 00000 20000 18000 22000 20000 30000 00000 00000 00000 30000 200001198603 39000 20000 00000 30000 19000 21000 24000 25000 25000 36000 20000 220001198611 16000 27000 30000 00000 19000 30000 23000 20000 10000 17000 40000 290001198612 10000 16000 10000 10000 00000 30000 24000 10000 20000 24000 26000 180001198613 01000 15000 12000 02000 00000 00000 21000 00000 01000 04000 10000 140001198621 20000 18000 20000 14000 16000 10000 00000 30000 00000 00000 10000 300001198622 10000 10000 25000 14000 18000 19000 27000 00000 20000 25000 15000 140001198623 25000 20000 29000 20000 19000 20000 26000 30000 00000 29000 10000 200001198631 30000 30000 30000 08000 23000 30000 24000 00000 00000 00000 40000 300001198632 29000 20000 00000 06000 16000 26000 21000 09000 00000 31000 00000 130001198633 18000 13000 14000 02000 09000 03000 10000 00000 00000 00000 18000 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 01105 01105 01105 00663 00552 00773 01105 00552 01105 00829 011051198602 00829 00000 00552 00497 00608 00552 00829 00000 00000 00000 00829 005521198603 01077 00552 00000 00829 00525 00580 00663 00691 00691 00994 00552 006081198611 00442 00746 00829 00000 00525 00829 00635 00552 00276 00470 01105 008011198612 00276 00442 00276 00276 00000 00829 00663 00276 00552 00663 00718 004971198613 00028 00414 00331 00055 00000 00000 00580 00000 00028 00110 00276 003871198621 00552 00497 00552 00387 00442 00276 00000 00829 00000 00000 00276 008291198622 00276 00276 00691 00387 00497 00525 00746 00000 00552 00691 00414 003871198623 00691 00552 00801 00552 00525 00552 00718 00829 00000 00801 00276 005521198631 00829 00829 00829 00221 00635 00829 00663 00000 00000 00000 01105 008291198632 00801 00552 00000 00166 00442 00718 00580 00249 00000 00856 00000 003591198633 00497 00359 00387 00055 00249 00083 00276 00000 00000 00000 00497 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633 119889

1198601 01250 02233 02211 01894 01618 01718 02066 01854 01023 02070 02120 02347 224041198602 01424 00664 01129 00954 01090 01150 01484 00500 00274 00582 01475 01249 119751198603 01991 01544 01007 01508 01311 01526 01722 01371 01064 01808 01621 01682 181551198611 01294 01542 01563 00593 01173 01606 01537 01094 00602 01181 01938 01663 157861198612 00915 01064 00878 00699 00504 01407 01334 00697 00753 01158 01356 01170 119361198613 00316 00647 00553 00240 00212 00230 00828 00183 00112 00296 00533 00655 048041198621 01085 01029 01082 00795 00883 00807 00629 01188 00273 00512 00885 01398 105671198622 00962 00947 01311 00855 01019 01164 01447 00487 00806 01242 01120 01116 124771198623 01521 01393 01621 01165 01205 01368 01635 01403 00376 01511 01215 01482 158951198631 01614 01602 01518 00802 01243 01561 01513 00561 00320 00695 01910 01665 150021198632 01319 01132 00593 00575 00890 01249 01196 00625 00217 01277 00654 01007 107341198633 00816 00679 00671 00315 00508 00399 00624 00252 00143 00309 00824 00359 05899119903 14507 14476 14136 10395 11656 14185 16015 10217 05964 12641 15651 15790




3

Impo

rtan

ce ra

nkin

g

Noncritical

Critical1

7

8

12


















Criteria 119889 119903 119889 + 119903 119889 minus 119903




5 Conclusions









1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00862 01542 01564 01822 01388 01211 01290 01815 01715 01637 01355 014861198602 00982 00459 00799 00917 00935 00810 00927 00490 00459 00461 00943 007911198603 01372 01066 00712 01451 01125 01076 01075 01342 01784 01430 01036 010651198611 00892 01065 01105 00570 01007 01132 00960 01071 01009 00934 01238 010531198612 00631 00735 00621 00673 00432 00992 00833 00682 01263 00916 00866 007411198613 00218 00447 00391 00230 00182 00162 00517 00179 00188 00234 00341 004151198621 00748 00711 00765 00765 00757 00569 00393 01163 00458 00405 00566 008851198622 00663 00654 00927 00822 00874 00821 00904 00477 01352 00983 00716 007071198623 01048 00963 01147 01121 01034 00965 01021 01374 00630 01195 00776 009381198631 01112 01106 01074 00771 01066 01101 00945 00549 00537 00549 01220 010541198632 00909 00782 00420 00554 00764 00880 00747 00612 00364 01011 00418 006381198633 00562 00469 00474 00303 00436 00281 00390 00247 00240 00245 00527 00227


1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 014691198602 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 007491198603 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 012381198611 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 009801198612 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 007661198613 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 002851198621 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 006871198622 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 008381198623 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 010311198631 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 009061198632 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 006661198633 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386




























maintenance (C2)






Acknowledgments


References





















































1 Introduction













2 Literature Review












Potholeimages


Vehicle stationary


optics

Center





Center

RSE

company

OBU


Pothole information

Potholeinformation


or WAVE



severity)

or

















Input image

Binarization by HST

Segmented images



Candidaterefinement




Candidate region




119862V

=

1 if 119878 (1198721015840119888) gt 119879119904 Com (1198721015840


0 otherwise

(1)







com (1198721015840119888) =1198972

4120587119860 (2)



1198621015840

V =










OHI (ℎ119888 ℎ119887) =

119899

sum

119894=0

min (oh119894119888 oh119894119887) (4)









119901

=


1198881015840 ℎ119887) gt 119879119900 OHI (ℎ1015840

1198881015840 ℎ1015840



(5)




1198881015840 ℎ1015840





proposed method





ℎ(threshold for





119879ℎ


image119879std1015840 10

119879119904 512 119879ave 00119879com 005 119879

119900087

119879std 8 1198791199001015840 085





















(a) Asphalt images

(b) Concrete images







6 Conclusions





(a) Original images

(b) Close-up images



(a) Bright images

(b) Dark images








Acknowledgment


References































Editorial Board










Contents















1 Introduction


















References
















1 Introduction













1198881 1198882 119888


1 1198882 119888

119899are


1 V2 V

119896 in



Custo

mer

-veh

icle

assig

nmen

t

Opt

imiz

ed as

signm

ent

CV

c1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

CV

c1c2

cn

12

k

middot



Fitness = min119899

sum

119894=0

119899

sum

119895=0

119896

sum

V=1119889119894119895119883

V119894119895+ 1198891198990119883

V1198990

119894 = 119895 (1)

119899

sum

119894=0

119899

sum

119895=0

119883

V119894119895119903119894le 119876V 119894 = 119895 V isin 119881 (2)

119883

V119894119895

=


0 otherwise

(3)


119894119895is the









1 1198882 119888

119899 to










119894= 119883

1198941 119883

119894119871





119894119871 whereas


119871


119881

new119894119895


times (119866119895minus 119883119894119895)

119883

new119894119895= 119883119894119895+ 119881

new119894119895

(4)



1

Depot

310

8

2

95

7

6

4

Opt

imiz

ed sc

hedu

le

Opt

imiz

ed as

signm

ent

1

Depot

72

8

10

95

3

6

4

7

Depot

310

8

5

92

1

6

4

CV

c1c2

cn

12

k

middot




119883

119881

119894119883

119881

1198941119883

119881


119881

119894119899119883

119881

119894119899+1119883

119881

119894119899+2sdot sdot sdot 119883

119881

119894119899+119896minus1











119894= 119883

119881

1198941 119883

119881

119894119899 119883

119881

119894119899+119896minus1


1Cus2 Cus

119899Veh1Veh2 Veh

119896minus1) For

















119863

1 119883

119863

2) is shown



Key2 13 08 24 19 02 12 21

02 08 12 13 19 2 21 24Key


Vehicle assignment

Visit order

Veh 1

Veh1

Veh1 Veh2

Veh2

Cus1

Cus1

Cus1

Veh 2

Cus2

Cus2

Cus2

Veh 3

Cus3

Cus3

Cus3

Cus4

Cus4

Cus4

Cus5

Cus5

Cus5

Cus6

Cus6

Cus6

XiV

XiV



X119863 119883

119863

1119883

119863

2






119862 119910119862)) can be yielded by

119909119862=

sum

119896

119894=1119909

V119894+ 119909119874

119896 + 1

119910119862=

sum

119896

119894=1119910

V119894+ 119910119874

119896 + 1

(5)

F

O

DE

G

HA

I

C

J

B

K

Route-1

Route-2

Route-3


F

O

DE

G

HA

I

C

J

B

K

dE

dF

dD

dC

dB

dA

CC


In (5) 119909119862and 119910





119874are the




F

O

DE

B

C

JK

G

H

I

A

(a) 1198891 = 119874119861 + 119861119866minus 119874119866

F

O

DE

B

C

JK

G

H

I

A

(b) 1198892 = 119866119861 + 119861119867minus 119866119867

F

O

DE

C

J

A

K

G

H

IB

(c) 1198893 = 119867119861 + 119861119868 minus 119867119868

F

O

DE

B

C

J

A

K

G

H

I

(d) 1198894 = 119868119861 + 119861119874minus 119868119874








(3) Outer layer PSO



layer PSO





Start




End

Yes Yes

NoNo

YesNo


Initialize VVX

V

Update VVX

V

Initialize VDX

D

Update VDX

D

search(XV)

Fitness(X ) lt

Fitness(XV)

Update(SA)

Fitness( )

Updateand

Pass XD

to inner layer PSO

Fitness(XD) =

Fitness( )= XLSV

GVbest

XVnew

PVbest

XVnew X

Vnew

Updateand

GVbest

PVbest

GVbest

LSV

XVLS = local




119899









A-n54-k9 53 + 53 53 + 8

A-n64-k8 63 + 63 63 + 7







1= 2 and 119888

2= 1 and the


1= 1198882= 2 and 100



BKStimes 100 (6)


Avg Dev () =sum

119899

119894=1DEV119894

119899

(7)




1) RBI (LS

2)




0102030405060708090

Aver

age d

evia

tion

()

A-n3

2-k5

A-n3

3-k5

A-n3

6-k5

A-n4

5-k6

A-n4

5-k7

A-n5

5-k9

A-n6

0-k9

A-n6

2-k8

A-n6

4-k9

Aver

age

wo LSLS1

LS2LS3




Imp() =Fitness


Fitness119908119900

times 100 (8)





Fitn

ess

950

900

850

800

750

700

Iterations

0

50

100

150

200

250

300

350

400

450

500

550

600

650

700

750

800

850

900

950

1000

(a)

Fitn

ess

830

810

790

770

750

730

710

690

670

650

Iterations

0 5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

(b)


(a) (b)






(a) (b)



5 Conclusions






Acknowledgment


References




































1 Introduction



























HMM correction

Vehicle V1

Vehicle V2

Vehicle Vn



based on HMM

Entireroutes

Routerecommendation


Vehicle Vi

A set ofOutput

Input

RS(Tk Roadi)

RouteT119896



upcomingroutes


Rm1Rm2

Rm3

Rm4

Rm5

Rm1x

Rm2y

Rm3x Rm4x

Rm5y

Y

X0


Ri

Rj

Rk

Rix

Rjy

Rkx

Rky

Y

X0



119860119899

997888rarr 119861 = 1198771119909 (1198771119910)


(1)





= (1 3) (1 4) (3 4) (3 1)

(2)






Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd


Sichuan RdJinmao St

Longpan Rd

Jianning Rd

0 1 2 3 4 5 6 7 80

1

2

3

4

5

6

Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd

Central Rd

Xufu Alley

Sichuan Rd

Jinmao St

Longpan Rd

Jianning Rd

X

Y












1199051 lt (1 3) (1 4) (3 4) (3 1) gt

1199052 lt (3 1) (3 4) (1 4) (1 3) gt

1199053 lt (0 3) (1 3) (1 5) (4 5) gt

1199054 lt (0 3) (0 0) (0 4) (4 1) gt

1199055 lt (2 0) (2 1) (3 1) (3 2) (4 2) gt

1199051 lt (1 3) (1 4) (3 4) (3 1) gt

(3)














997888997888997888997888rarr119860119895119861119895 = (119909119895119899minus1199091198951 119910119895119899minus



cos ⟨ 119886119894 119886119895⟩ =

119886119894 sdot 119886119895

1003816100381610038161003816 1198861198941003816100381610038161003816 sdot10038161003816100381610038161003816119886119895

10038161003816100381610038161003816

= ((119909119894119899 minus 1199091198941) sdot (119910119894119899 minus 1199101198941)

+ (119909119895119899 minus 1199091198951) sdot (119910119895119899 minus 1199101198951))


2

sdotradic(119909119895119899 minus 1199091198951)2

+ (119910119895119899 minus 1199101198951)2

)

minus1

(4)


1





120587119894 =

Count (119904119863119894

) + 120582Count (119904New119863119894

)

sum119899

119895=1[Count (119904119863

119895

) + 120582Count (119904New119863119895

)]

(5)




1198961= 1199001198962)






119894

)

and Count(119904New119863119894



119875 (119904119894 | 119904119894minus1)

=

Count (119904119863119894minus1

119904119863119894

) + 120582Count (119904New119863119894minus1

119904New119863119894

) + 1

Count (119904119863119894minus1

) + 120582Count (119904New119863119894minus1

) + 119898

(6)


119904119863119894


119904New119863119894

)



119875 (V119895 | 119904119894)

=

Count (119904119863119894minus1

V119863119894

) + 120582Count (119904New119863119894minus1

VNew119863119894

) + 1

Count (119904119863119894

) + 120582Count (119904New119863119894

) + 119899

(7)


V119863119894


VNew119863119894

)






sequence

Viterbialgorithm










119901119894(1199091) 119901119894(119909

2) 119901119894(119909



1) 119901119894(119910

2) 119901119894(119910

119899) (1198632 notin 119881) and the











(b) Suppose that 1198632 = 119901119894(1199101) 119901119894(119910

2) 119901119894119896 then


(c) Suppose that 1198632 = 1199011198941 119901119894(1199102) 119901119894(119910

119899) then



1) 119901119894(119910

2) 119901119894(119910

119899) and 1199101 = 1 119910119899


1)

119901119894(1199102) 119901119894(119910





















119875119894 =sum119899

119896=1119863(119877119896 119862119877)

sum119899

119905=1Dist 1003816100381610038161003816119877119905

1003816100381610038161003816

times 100 (8)



1198751 =Repeat (1198771 1198771) + Repeat (1198773 1198771)

Dist 100381610038161003816100381611987711003816100381610038161003816 + Dist 10038161003816100381610038161198773

1003816100381610038161003816

times 100

=6 + 1

6 + 7times 100 = 5385

(9)












(a) (b)


100908070605040302010

01009080706050403020100

Trip completed ()

Cor

rect

pre

dict

ion

()


Cor

rect

pre

dict

ion

()

100908070605040302010

01009080706050403020100

Trip completed ()


Cor

rect

pre

dict

ion

()


100908070605040302010

0


120582 = 05

120582 = 02


Cor

rect

pre

dict

ion

()

100908070605040302010

0



120582 = 05

120582 = 02







90

80

70

60

50

40

30

20

10

0Other time periods

Cor

rect

pre

dict

ion

()

Time of day







7 Conclusion






Acknowledgments


References

























1 Introduction



















1198811198732

1198811198733

1198811198734

1198811198731

(1198811198731 1198811198731) (119881119873

1 1198811198732) (119881119873

1 1198811198733) (119881119873

1 1198811198734)

1198811198732

(1198811198732 1198811198731) (119881119873

2 1198811198732) (119881119873

2 1198811198733) (119881119873

2 1198811198734)

1198811198733

(1198811198733 1198811198731) (119881119873

3 1198811198732) (119881119873

3 1198811198733) (119881119873

3 1198811198734)

1198811198734

(1198811198734 1198811198731) (119881119873

4 1198811198732) (119881119873

4 1198811198733) (119881119873

4 1198811198734)



1 1199051⟩ rarr ⟨119873


119899 119905119899⟩)


1rarr 119873


119899)




119899) where

tr(119896+1)



tr(119896+1)


119896




1rarr 119873


119873119896minus1


1








Subregion 1

Subregion 2

Subregion 3

Subregion 4

Node 1Node 4

Node 2Node 3

Virtual node 4


Virtual node 1

Virtual node 4


Virtual node 1

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Physical topology

Logical topology


Link 2

Link 5

Link 1

Path 1 Path 2

Link 3

Link 4

Path 3 Path 4


Link 1


Link 5

Path 1

Path 2

Path 3

Path 4

Path 1Link 1

Link 3

Link 4

Path 2


Path 3Link 2

Link 3

Link 4 Path 2

Link 3Link 2


Path 1 Path 3

Path 4

Link 4

Path 2

(b) Index






[[[[[[[[

[

11990911

11990912


1199091119908

11990921

11990922


1199092119908


119909119879minus11

119909119879minus12


119909119879minus1119908

1199091198791

1199091198792


119909119879119908

]]]]]]]]

]

(1)








119885119871 119884119898

(119898 = 1 119871) (2)


119871 119884119898(119898 = 1 119871) in every scale is





119871 119898(119898 = 1 119871) at all scales



SPE119894=

119882

sum

119895=1

(119909119894119895minus 119909119894119895)2

(3)






1205852

120572= 1206011

[[

[

119888120572radic21206012ℎ2

0

1206011

+ 1 +1206012ℎ0(ℎ0minus 1)

1206012

1

]]

]

1ℎ0

(4)


1120601331206012

2 120601119894= sum119882

119895=119903+1120582119894

119895 119894 = 1 2 3 120582

119895is


120572is the 1 minus 120572





Transform

Physical topology

Logical topology

Taxi GPSdata

Traffic flowdata


transformPCA




Judge

Anomalousposition



119905

Thus

119876119905+1= 120573119876119905+ (1 minus 120573)119876

119905 (5)




120595 = 120583119904+ 119871 times 120590

119904radic

120573

(2 minus 120573) 119879 (6)
























119889119862119903 where 119862





GPS data



















1 432012 5927 6297 83172 632012 6418 6452 75863 732012 5344 7020 8849

32 1152012 6374 7270 8237

74 3182012 4728 7737 7888Average 119877 value 5633 6329 7462

050

100150200250300350400450500

Traffi

c flow


t

700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

20

40

60

80

100

120

140

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

50

100

150

200

250

300

350

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

10

20

30

40

50

60

70

80

()


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

845

ndash900

900

ndash915

915

ndash930











0

2000

4000

6000

8000

10000

12000

14000

16000


Proc

essin

g tim

e (m

s)



4 Conclusions












Acknowledgments



References











































1 Introduction










2 Literature Review









Organization


[35ndash43]

Project base


[36 37 40 41 43]

Systemtechnology


[35ndash43]

Consultants


[35ndash37 39]

Externalenvironment


[38 42]



3 Methodology








Organization





Project base





Systemtechnology






Consultant





Externalenvironment









Normalized Z(X)




each factor







DEMATEL

ANP





Z =

[[[[[[[

[

1199111111991112sdot sdot sdot 119911

1119899

1199112111991122sdot sdot sdot 119911

2119899

11991111989911199111198992sdot sdot sdot 119911

119899119899

]]]]]]]

]

(1)


120582 =

1

max1le119894le119899sum119899

119895=1119885119894119895

(119894 119895 = 1 2 119899)

X = 120582 sdot Z(2)


lim119870rarrinfin

X119870 = 0


(3)



Z =119860

119861

119862

119863

((

(

119860

0

3

3

3

119861

2

0

1

2

119862

2

2

0

2

119863

2

1

2

0

))

)

(4)


X =119860

119861

119862

119863

((

(

119860

0000

0337

0326

0337

119861

0233

0000

0116

0198

119862

0279

0198

0000

0198

119863

0233

0116

0244

0000

))

)

T =

119860

119861

119862

119863

(

119860

0628

0817

0839

0876

119861

0580

0356

0483

0559

119862

0691

0593

0449

0637

119863

0615

0493

0605

0424

)

119889

2513

2259

2377

2497

119903 3159 1979 2370 2137

(5)





Table 3


119860 2513 3159 5673 1 minus06462119861 2259 1979 4238 4 02796119862 2377 2370 4746 2 00068119863 2496 2137 4633 3 03598

Table 4

119860 119861 119862 119863

119860 0199 0293 0291 0288119861 0259 0180 0250 0231119862 0266 0244 0190 0283119863 0277 0283 0269 0199





Table 5


Overallrankings


A B C DA 0628 0580 0691 0615B 0817 0256 0593 0493C 0839 0483 0449 0605D 0876 0559 0637 0424

T =

Figure 2

DA

C

Figure 3

4 Empirical Study



























43 Result Analysis






Aspects 119860 119861 119862 119863

119860 00000 20000 24000 20000119861 29000 00000 17000 10000119862 28000 10000 00000 21000119863 29000 17000 17000 00000


Aspects 119860 119861 119862 119863

119860 00000 02326 02791 02326119861 03372 00000 01977 01163119862 03256 01163 00000 02442119863 03372 01977 01977 00000


Aspects 119860 119861 119862 119863 119889

119860 06278 05803 06905 06146 25132119861 08166 03563 05933 04925 22587119862 08389 04832 04492 06052 23765119863 08761 05593 06366 04242 24963119903 31593 19791 23697 21365


Aspects 119889 119903 119889 + 119903 119889 minus 119903

119860 25132 31593 56725 minus06462119861 22587 19791 42378 02796119862 23765 23697 47461 00068119863 24963 21365 46328 03598


Aspects 119860 119861 119862 119863

119860 01987 02932 02914 02877119861 02585 01800 02504 02305119862 02655 02442 01896 02832119863 02773 02826 02686 01986


Aspects 119860 119861 119862 119863

119860 02662 02662 02662 02662119861 02312 02312 02312 02312119862 02464 02464 02464 02464119863 02562 02562 02562 02562




Overallrankings



Organizations(A)


(D)System

technology (C)

Project base (B)









Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 40000 40000 40000 24000 20000 28000 40000 20000 40000 30000 400001198602 30000 00000 20000 18000 22000 20000 30000 00000 00000 00000 30000 200001198603 39000 20000 00000 30000 19000 21000 24000 25000 25000 36000 20000 220001198611 16000 27000 30000 00000 19000 30000 23000 20000 10000 17000 40000 290001198612 10000 16000 10000 10000 00000 30000 24000 10000 20000 24000 26000 180001198613 01000 15000 12000 02000 00000 00000 21000 00000 01000 04000 10000 140001198621 20000 18000 20000 14000 16000 10000 00000 30000 00000 00000 10000 300001198622 10000 10000 25000 14000 18000 19000 27000 00000 20000 25000 15000 140001198623 25000 20000 29000 20000 19000 20000 26000 30000 00000 29000 10000 200001198631 30000 30000 30000 08000 23000 30000 24000 00000 00000 00000 40000 300001198632 29000 20000 00000 06000 16000 26000 21000 09000 00000 31000 00000 130001198633 18000 13000 14000 02000 09000 03000 10000 00000 00000 00000 18000 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 01105 01105 01105 00663 00552 00773 01105 00552 01105 00829 011051198602 00829 00000 00552 00497 00608 00552 00829 00000 00000 00000 00829 005521198603 01077 00552 00000 00829 00525 00580 00663 00691 00691 00994 00552 006081198611 00442 00746 00829 00000 00525 00829 00635 00552 00276 00470 01105 008011198612 00276 00442 00276 00276 00000 00829 00663 00276 00552 00663 00718 004971198613 00028 00414 00331 00055 00000 00000 00580 00000 00028 00110 00276 003871198621 00552 00497 00552 00387 00442 00276 00000 00829 00000 00000 00276 008291198622 00276 00276 00691 00387 00497 00525 00746 00000 00552 00691 00414 003871198623 00691 00552 00801 00552 00525 00552 00718 00829 00000 00801 00276 005521198631 00829 00829 00829 00221 00635 00829 00663 00000 00000 00000 01105 008291198632 00801 00552 00000 00166 00442 00718 00580 00249 00000 00856 00000 003591198633 00497 00359 00387 00055 00249 00083 00276 00000 00000 00000 00497 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633 119889

1198601 01250 02233 02211 01894 01618 01718 02066 01854 01023 02070 02120 02347 224041198602 01424 00664 01129 00954 01090 01150 01484 00500 00274 00582 01475 01249 119751198603 01991 01544 01007 01508 01311 01526 01722 01371 01064 01808 01621 01682 181551198611 01294 01542 01563 00593 01173 01606 01537 01094 00602 01181 01938 01663 157861198612 00915 01064 00878 00699 00504 01407 01334 00697 00753 01158 01356 01170 119361198613 00316 00647 00553 00240 00212 00230 00828 00183 00112 00296 00533 00655 048041198621 01085 01029 01082 00795 00883 00807 00629 01188 00273 00512 00885 01398 105671198622 00962 00947 01311 00855 01019 01164 01447 00487 00806 01242 01120 01116 124771198623 01521 01393 01621 01165 01205 01368 01635 01403 00376 01511 01215 01482 158951198631 01614 01602 01518 00802 01243 01561 01513 00561 00320 00695 01910 01665 150021198632 01319 01132 00593 00575 00890 01249 01196 00625 00217 01277 00654 01007 107341198633 00816 00679 00671 00315 00508 00399 00624 00252 00143 00309 00824 00359 05899119903 14507 14476 14136 10395 11656 14185 16015 10217 05964 12641 15651 15790




3

Impo

rtan

ce ra

nkin

g

Noncritical

Critical1

7

8

12


















Criteria 119889 119903 119889 + 119903 119889 minus 119903




5 Conclusions









1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00862 01542 01564 01822 01388 01211 01290 01815 01715 01637 01355 014861198602 00982 00459 00799 00917 00935 00810 00927 00490 00459 00461 00943 007911198603 01372 01066 00712 01451 01125 01076 01075 01342 01784 01430 01036 010651198611 00892 01065 01105 00570 01007 01132 00960 01071 01009 00934 01238 010531198612 00631 00735 00621 00673 00432 00992 00833 00682 01263 00916 00866 007411198613 00218 00447 00391 00230 00182 00162 00517 00179 00188 00234 00341 004151198621 00748 00711 00765 00765 00757 00569 00393 01163 00458 00405 00566 008851198622 00663 00654 00927 00822 00874 00821 00904 00477 01352 00983 00716 007071198623 01048 00963 01147 01121 01034 00965 01021 01374 00630 01195 00776 009381198631 01112 01106 01074 00771 01066 01101 00945 00549 00537 00549 01220 010541198632 00909 00782 00420 00554 00764 00880 00747 00612 00364 01011 00418 006381198633 00562 00469 00474 00303 00436 00281 00390 00247 00240 00245 00527 00227


1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 014691198602 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 007491198603 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 012381198611 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 009801198612 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 007661198613 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 002851198621 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 006871198622 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 008381198623 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 010311198631 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 009061198632 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 006661198633 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386




























maintenance (C2)






Acknowledgments


References





















































1 Introduction













2 Literature Review












Potholeimages


Vehicle stationary


optics

Center





Center

RSE

company

OBU


Pothole information

Potholeinformation


or WAVE



severity)

or

















Input image

Binarization by HST

Segmented images



Candidaterefinement




Candidate region




119862V

=

1 if 119878 (1198721015840119888) gt 119879119904 Com (1198721015840


0 otherwise

(1)







com (1198721015840119888) =1198972

4120587119860 (2)



1198621015840

V =










OHI (ℎ119888 ℎ119887) =

119899

sum

119894=0

min (oh119894119888 oh119894119887) (4)









119901

=


1198881015840 ℎ119887) gt 119879119900 OHI (ℎ1015840

1198881015840 ℎ1015840



(5)




1198881015840 ℎ1015840





proposed method





ℎ(threshold for





119879ℎ


image119879std1015840 10

119879119904 512 119879ave 00119879com 005 119879

119900087

119879std 8 1198791199001015840 085





















(a) Asphalt images

(b) Concrete images







6 Conclusions





(a) Original images

(b) Close-up images



(a) Bright images

(b) Dark images








Acknowledgment


References





































Contents















1 Introduction


















References
















1 Introduction













1198881 1198882 119888


1 1198882 119888

119899are


1 V2 V

119896 in



Custo

mer

-veh

icle

assig

nmen

t

Opt

imiz

ed as

signm

ent

CV

c1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

CV

c1c2

cn

12

k

middot



Fitness = min119899

sum

119894=0

119899

sum

119895=0

119896

sum

V=1119889119894119895119883

V119894119895+ 1198891198990119883

V1198990

119894 = 119895 (1)

119899

sum

119894=0

119899

sum

119895=0

119883

V119894119895119903119894le 119876V 119894 = 119895 V isin 119881 (2)

119883

V119894119895

=


0 otherwise

(3)


119894119895is the









1 1198882 119888

119899 to










119894= 119883

1198941 119883

119894119871





119894119871 whereas


119871


119881

new119894119895


times (119866119895minus 119883119894119895)

119883

new119894119895= 119883119894119895+ 119881

new119894119895

(4)



1

Depot

310

8

2

95

7

6

4

Opt

imiz

ed sc

hedu

le

Opt

imiz

ed as

signm

ent

1

Depot

72

8

10

95

3

6

4

7

Depot

310

8

5

92

1

6

4

CV

c1c2

cn

12

k

middot




119883

119881

119894119883

119881

1198941119883

119881


119881

119894119899119883

119881

119894119899+1119883

119881

119894119899+2sdot sdot sdot 119883

119881

119894119899+119896minus1











119894= 119883

119881

1198941 119883

119881

119894119899 119883

119881

119894119899+119896minus1


1Cus2 Cus

119899Veh1Veh2 Veh

119896minus1) For

















119863

1 119883

119863

2) is shown



Key2 13 08 24 19 02 12 21

02 08 12 13 19 2 21 24Key


Vehicle assignment

Visit order

Veh 1

Veh1

Veh1 Veh2

Veh2

Cus1

Cus1

Cus1

Veh 2

Cus2

Cus2

Cus2

Veh 3

Cus3

Cus3

Cus3

Cus4

Cus4

Cus4

Cus5

Cus5

Cus5

Cus6

Cus6

Cus6

XiV

XiV



X119863 119883

119863

1119883

119863

2






119862 119910119862)) can be yielded by

119909119862=

sum

119896

119894=1119909

V119894+ 119909119874

119896 + 1

119910119862=

sum

119896

119894=1119910

V119894+ 119910119874

119896 + 1

(5)

F

O

DE

G

HA

I

C

J

B

K

Route-1

Route-2

Route-3


F

O

DE

G

HA

I

C

J

B

K

dE

dF

dD

dC

dB

dA

CC


In (5) 119909119862and 119910





119874are the




F

O

DE

B

C

JK

G

H

I

A

(a) 1198891 = 119874119861 + 119861119866minus 119874119866

F

O

DE

B

C

JK

G

H

I

A

(b) 1198892 = 119866119861 + 119861119867minus 119866119867

F

O

DE

C

J

A

K

G

H

IB

(c) 1198893 = 119867119861 + 119861119868 minus 119867119868

F

O

DE

B

C

J

A

K

G

H

I

(d) 1198894 = 119868119861 + 119861119874minus 119868119874








(3) Outer layer PSO



layer PSO





Start




End

Yes Yes

NoNo

YesNo


Initialize VVX

V

Update VVX

V

Initialize VDX

D

Update VDX

D

search(XV)

Fitness(X ) lt

Fitness(XV)

Update(SA)

Fitness( )

Updateand

Pass XD

to inner layer PSO

Fitness(XD) =

Fitness( )= XLSV

GVbest

XVnew

PVbest

XVnew X

Vnew

Updateand

GVbest

PVbest

GVbest

LSV

XVLS = local




119899









A-n54-k9 53 + 53 53 + 8

A-n64-k8 63 + 63 63 + 7







1= 2 and 119888

2= 1 and the


1= 1198882= 2 and 100



BKStimes 100 (6)


Avg Dev () =sum

119899

119894=1DEV119894

119899

(7)




1) RBI (LS

2)




0102030405060708090

Aver

age d

evia

tion

()

A-n3

2-k5

A-n3

3-k5

A-n3

6-k5

A-n4

5-k6

A-n4

5-k7

A-n5

5-k9

A-n6

0-k9

A-n6

2-k8

A-n6

4-k9

Aver

age

wo LSLS1

LS2LS3




Imp() =Fitness


Fitness119908119900

times 100 (8)





Fitn

ess

950

900

850

800

750

700

Iterations

0

50

100

150

200

250

300

350

400

450

500

550

600

650

700

750

800

850

900

950

1000

(a)

Fitn

ess

830

810

790

770

750

730

710

690

670

650

Iterations

0 5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

(b)


(a) (b)






(a) (b)



5 Conclusions






Acknowledgment


References




































1 Introduction



























HMM correction

Vehicle V1

Vehicle V2

Vehicle Vn



based on HMM

Entireroutes

Routerecommendation


Vehicle Vi

A set ofOutput

Input

RS(Tk Roadi)

RouteT119896



upcomingroutes


Rm1Rm2

Rm3

Rm4

Rm5

Rm1x

Rm2y

Rm3x Rm4x

Rm5y

Y

X0


Ri

Rj

Rk

Rix

Rjy

Rkx

Rky

Y

X0



119860119899

997888rarr 119861 = 1198771119909 (1198771119910)


(1)





= (1 3) (1 4) (3 4) (3 1)

(2)






Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd


Sichuan RdJinmao St

Longpan Rd

Jianning Rd

0 1 2 3 4 5 6 7 80

1

2

3

4

5

6

Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd

Central Rd

Xufu Alley

Sichuan Rd

Jinmao St

Longpan Rd

Jianning Rd

X

Y












1199051 lt (1 3) (1 4) (3 4) (3 1) gt

1199052 lt (3 1) (3 4) (1 4) (1 3) gt

1199053 lt (0 3) (1 3) (1 5) (4 5) gt

1199054 lt (0 3) (0 0) (0 4) (4 1) gt

1199055 lt (2 0) (2 1) (3 1) (3 2) (4 2) gt

1199051 lt (1 3) (1 4) (3 4) (3 1) gt

(3)














997888997888997888997888rarr119860119895119861119895 = (119909119895119899minus1199091198951 119910119895119899minus



cos ⟨ 119886119894 119886119895⟩ =

119886119894 sdot 119886119895

1003816100381610038161003816 1198861198941003816100381610038161003816 sdot10038161003816100381610038161003816119886119895

10038161003816100381610038161003816

= ((119909119894119899 minus 1199091198941) sdot (119910119894119899 minus 1199101198941)

+ (119909119895119899 minus 1199091198951) sdot (119910119895119899 minus 1199101198951))


2

sdotradic(119909119895119899 minus 1199091198951)2

+ (119910119895119899 minus 1199101198951)2

)

minus1

(4)


1





120587119894 =

Count (119904119863119894

) + 120582Count (119904New119863119894

)

sum119899

119895=1[Count (119904119863

119895

) + 120582Count (119904New119863119895

)]

(5)




1198961= 1199001198962)






119894

)

and Count(119904New119863119894



119875 (119904119894 | 119904119894minus1)

=

Count (119904119863119894minus1

119904119863119894

) + 120582Count (119904New119863119894minus1

119904New119863119894

) + 1

Count (119904119863119894minus1

) + 120582Count (119904New119863119894minus1

) + 119898

(6)


119904119863119894


119904New119863119894

)



119875 (V119895 | 119904119894)

=

Count (119904119863119894minus1

V119863119894

) + 120582Count (119904New119863119894minus1

VNew119863119894

) + 1

Count (119904119863119894

) + 120582Count (119904New119863119894

) + 119899

(7)


V119863119894


VNew119863119894

)






sequence

Viterbialgorithm










119901119894(1199091) 119901119894(119909

2) 119901119894(119909



1) 119901119894(119910

2) 119901119894(119910

119899) (1198632 notin 119881) and the











(b) Suppose that 1198632 = 119901119894(1199101) 119901119894(119910

2) 119901119894119896 then


(c) Suppose that 1198632 = 1199011198941 119901119894(1199102) 119901119894(119910

119899) then



1) 119901119894(119910

2) 119901119894(119910

119899) and 1199101 = 1 119910119899


1)

119901119894(1199102) 119901119894(119910





















119875119894 =sum119899

119896=1119863(119877119896 119862119877)

sum119899

119905=1Dist 1003816100381610038161003816119877119905

1003816100381610038161003816

times 100 (8)



1198751 =Repeat (1198771 1198771) + Repeat (1198773 1198771)

Dist 100381610038161003816100381611987711003816100381610038161003816 + Dist 10038161003816100381610038161198773

1003816100381610038161003816

times 100

=6 + 1

6 + 7times 100 = 5385

(9)












(a) (b)


100908070605040302010

01009080706050403020100

Trip completed ()

Cor

rect

pre

dict

ion

()


Cor

rect

pre

dict

ion

()

100908070605040302010

01009080706050403020100

Trip completed ()


Cor

rect

pre

dict

ion

()


100908070605040302010

0


120582 = 05

120582 = 02


Cor

rect

pre

dict

ion

()

100908070605040302010

0



120582 = 05

120582 = 02







90

80

70

60

50

40

30

20

10

0Other time periods

Cor

rect

pre

dict

ion

()

Time of day







7 Conclusion






Acknowledgments


References

























1 Introduction



















1198811198732

1198811198733

1198811198734

1198811198731

(1198811198731 1198811198731) (119881119873

1 1198811198732) (119881119873

1 1198811198733) (119881119873

1 1198811198734)

1198811198732

(1198811198732 1198811198731) (119881119873

2 1198811198732) (119881119873

2 1198811198733) (119881119873

2 1198811198734)

1198811198733

(1198811198733 1198811198731) (119881119873

3 1198811198732) (119881119873

3 1198811198733) (119881119873

3 1198811198734)

1198811198734

(1198811198734 1198811198731) (119881119873

4 1198811198732) (119881119873

4 1198811198733) (119881119873

4 1198811198734)



1 1199051⟩ rarr ⟨119873


119899 119905119899⟩)


1rarr 119873


119899)




119899) where

tr(119896+1)



tr(119896+1)


119896




1rarr 119873


119873119896minus1


1








Subregion 1

Subregion 2

Subregion 3

Subregion 4

Node 1Node 4

Node 2Node 3

Virtual node 4


Virtual node 1

Virtual node 4


Virtual node 1

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Physical topology

Logical topology


Link 2

Link 5

Link 1

Path 1 Path 2

Link 3

Link 4

Path 3 Path 4


Link 1


Link 5

Path 1

Path 2

Path 3

Path 4

Path 1Link 1

Link 3

Link 4

Path 2


Path 3Link 2

Link 3

Link 4 Path 2

Link 3Link 2


Path 1 Path 3

Path 4

Link 4

Path 2

(b) Index






[[[[[[[[

[

11990911

11990912


1199091119908

11990921

11990922


1199092119908


119909119879minus11

119909119879minus12


119909119879minus1119908

1199091198791

1199091198792


119909119879119908

]]]]]]]]

]

(1)








119885119871 119884119898

(119898 = 1 119871) (2)


119871 119884119898(119898 = 1 119871) in every scale is





119871 119898(119898 = 1 119871) at all scales



SPE119894=

119882

sum

119895=1

(119909119894119895minus 119909119894119895)2

(3)






1205852

120572= 1206011

[[

[

119888120572radic21206012ℎ2

0

1206011

+ 1 +1206012ℎ0(ℎ0minus 1)

1206012

1

]]

]

1ℎ0

(4)


1120601331206012

2 120601119894= sum119882

119895=119903+1120582119894

119895 119894 = 1 2 3 120582

119895is


120572is the 1 minus 120572





Transform

Physical topology

Logical topology

Taxi GPSdata

Traffic flowdata


transformPCA




Judge

Anomalousposition



119905

Thus

119876119905+1= 120573119876119905+ (1 minus 120573)119876

119905 (5)




120595 = 120583119904+ 119871 times 120590

119904radic

120573

(2 minus 120573) 119879 (6)
























119889119862119903 where 119862





GPS data



















1 432012 5927 6297 83172 632012 6418 6452 75863 732012 5344 7020 8849

32 1152012 6374 7270 8237

74 3182012 4728 7737 7888Average 119877 value 5633 6329 7462

050

100150200250300350400450500

Traffi

c flow


t

700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

20

40

60

80

100

120

140

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

50

100

150

200

250

300

350

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

10

20

30

40

50

60

70

80

()


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

845

ndash900

900

ndash915

915

ndash930











0

2000

4000

6000

8000

10000

12000

14000

16000


Proc

essin

g tim

e (m

s)



4 Conclusions












Acknowledgments



References











































1 Introduction










2 Literature Review









Organization


[35ndash43]

Project base


[36 37 40 41 43]

Systemtechnology


[35ndash43]

Consultants


[35ndash37 39]

Externalenvironment


[38 42]



3 Methodology








Organization





Project base





Systemtechnology






Consultant





Externalenvironment









Normalized Z(X)




each factor







DEMATEL

ANP





Z =

[[[[[[[

[

1199111111991112sdot sdot sdot 119911

1119899

1199112111991122sdot sdot sdot 119911

2119899

11991111989911199111198992sdot sdot sdot 119911

119899119899

]]]]]]]

]

(1)


120582 =

1

max1le119894le119899sum119899

119895=1119885119894119895

(119894 119895 = 1 2 119899)

X = 120582 sdot Z(2)


lim119870rarrinfin

X119870 = 0


(3)



Z =119860

119861

119862

119863

((

(

119860

0

3

3

3

119861

2

0

1

2

119862

2

2

0

2

119863

2

1

2

0

))

)

(4)


X =119860

119861

119862

119863

((

(

119860

0000

0337

0326

0337

119861

0233

0000

0116

0198

119862

0279

0198

0000

0198

119863

0233

0116

0244

0000

))

)

T =

119860

119861

119862

119863

(

119860

0628

0817

0839

0876

119861

0580

0356

0483

0559

119862

0691

0593

0449

0637

119863

0615

0493

0605

0424

)

119889

2513

2259

2377

2497

119903 3159 1979 2370 2137

(5)





Table 3


119860 2513 3159 5673 1 minus06462119861 2259 1979 4238 4 02796119862 2377 2370 4746 2 00068119863 2496 2137 4633 3 03598

Table 4

119860 119861 119862 119863

119860 0199 0293 0291 0288119861 0259 0180 0250 0231119862 0266 0244 0190 0283119863 0277 0283 0269 0199





Table 5


Overallrankings


A B C DA 0628 0580 0691 0615B 0817 0256 0593 0493C 0839 0483 0449 0605D 0876 0559 0637 0424

T =

Figure 2

DA

C

Figure 3

4 Empirical Study



























43 Result Analysis






Aspects 119860 119861 119862 119863

119860 00000 20000 24000 20000119861 29000 00000 17000 10000119862 28000 10000 00000 21000119863 29000 17000 17000 00000


Aspects 119860 119861 119862 119863

119860 00000 02326 02791 02326119861 03372 00000 01977 01163119862 03256 01163 00000 02442119863 03372 01977 01977 00000


Aspects 119860 119861 119862 119863 119889

119860 06278 05803 06905 06146 25132119861 08166 03563 05933 04925 22587119862 08389 04832 04492 06052 23765119863 08761 05593 06366 04242 24963119903 31593 19791 23697 21365


Aspects 119889 119903 119889 + 119903 119889 minus 119903

119860 25132 31593 56725 minus06462119861 22587 19791 42378 02796119862 23765 23697 47461 00068119863 24963 21365 46328 03598


Aspects 119860 119861 119862 119863

119860 01987 02932 02914 02877119861 02585 01800 02504 02305119862 02655 02442 01896 02832119863 02773 02826 02686 01986


Aspects 119860 119861 119862 119863

119860 02662 02662 02662 02662119861 02312 02312 02312 02312119862 02464 02464 02464 02464119863 02562 02562 02562 02562




Overallrankings



Organizations(A)


(D)System

technology (C)

Project base (B)









Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 40000 40000 40000 24000 20000 28000 40000 20000 40000 30000 400001198602 30000 00000 20000 18000 22000 20000 30000 00000 00000 00000 30000 200001198603 39000 20000 00000 30000 19000 21000 24000 25000 25000 36000 20000 220001198611 16000 27000 30000 00000 19000 30000 23000 20000 10000 17000 40000 290001198612 10000 16000 10000 10000 00000 30000 24000 10000 20000 24000 26000 180001198613 01000 15000 12000 02000 00000 00000 21000 00000 01000 04000 10000 140001198621 20000 18000 20000 14000 16000 10000 00000 30000 00000 00000 10000 300001198622 10000 10000 25000 14000 18000 19000 27000 00000 20000 25000 15000 140001198623 25000 20000 29000 20000 19000 20000 26000 30000 00000 29000 10000 200001198631 30000 30000 30000 08000 23000 30000 24000 00000 00000 00000 40000 300001198632 29000 20000 00000 06000 16000 26000 21000 09000 00000 31000 00000 130001198633 18000 13000 14000 02000 09000 03000 10000 00000 00000 00000 18000 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 01105 01105 01105 00663 00552 00773 01105 00552 01105 00829 011051198602 00829 00000 00552 00497 00608 00552 00829 00000 00000 00000 00829 005521198603 01077 00552 00000 00829 00525 00580 00663 00691 00691 00994 00552 006081198611 00442 00746 00829 00000 00525 00829 00635 00552 00276 00470 01105 008011198612 00276 00442 00276 00276 00000 00829 00663 00276 00552 00663 00718 004971198613 00028 00414 00331 00055 00000 00000 00580 00000 00028 00110 00276 003871198621 00552 00497 00552 00387 00442 00276 00000 00829 00000 00000 00276 008291198622 00276 00276 00691 00387 00497 00525 00746 00000 00552 00691 00414 003871198623 00691 00552 00801 00552 00525 00552 00718 00829 00000 00801 00276 005521198631 00829 00829 00829 00221 00635 00829 00663 00000 00000 00000 01105 008291198632 00801 00552 00000 00166 00442 00718 00580 00249 00000 00856 00000 003591198633 00497 00359 00387 00055 00249 00083 00276 00000 00000 00000 00497 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633 119889

1198601 01250 02233 02211 01894 01618 01718 02066 01854 01023 02070 02120 02347 224041198602 01424 00664 01129 00954 01090 01150 01484 00500 00274 00582 01475 01249 119751198603 01991 01544 01007 01508 01311 01526 01722 01371 01064 01808 01621 01682 181551198611 01294 01542 01563 00593 01173 01606 01537 01094 00602 01181 01938 01663 157861198612 00915 01064 00878 00699 00504 01407 01334 00697 00753 01158 01356 01170 119361198613 00316 00647 00553 00240 00212 00230 00828 00183 00112 00296 00533 00655 048041198621 01085 01029 01082 00795 00883 00807 00629 01188 00273 00512 00885 01398 105671198622 00962 00947 01311 00855 01019 01164 01447 00487 00806 01242 01120 01116 124771198623 01521 01393 01621 01165 01205 01368 01635 01403 00376 01511 01215 01482 158951198631 01614 01602 01518 00802 01243 01561 01513 00561 00320 00695 01910 01665 150021198632 01319 01132 00593 00575 00890 01249 01196 00625 00217 01277 00654 01007 107341198633 00816 00679 00671 00315 00508 00399 00624 00252 00143 00309 00824 00359 05899119903 14507 14476 14136 10395 11656 14185 16015 10217 05964 12641 15651 15790




3

Impo

rtan

ce ra

nkin

g

Noncritical

Critical1

7

8

12


















Criteria 119889 119903 119889 + 119903 119889 minus 119903




5 Conclusions









1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00862 01542 01564 01822 01388 01211 01290 01815 01715 01637 01355 014861198602 00982 00459 00799 00917 00935 00810 00927 00490 00459 00461 00943 007911198603 01372 01066 00712 01451 01125 01076 01075 01342 01784 01430 01036 010651198611 00892 01065 01105 00570 01007 01132 00960 01071 01009 00934 01238 010531198612 00631 00735 00621 00673 00432 00992 00833 00682 01263 00916 00866 007411198613 00218 00447 00391 00230 00182 00162 00517 00179 00188 00234 00341 004151198621 00748 00711 00765 00765 00757 00569 00393 01163 00458 00405 00566 008851198622 00663 00654 00927 00822 00874 00821 00904 00477 01352 00983 00716 007071198623 01048 00963 01147 01121 01034 00965 01021 01374 00630 01195 00776 009381198631 01112 01106 01074 00771 01066 01101 00945 00549 00537 00549 01220 010541198632 00909 00782 00420 00554 00764 00880 00747 00612 00364 01011 00418 006381198633 00562 00469 00474 00303 00436 00281 00390 00247 00240 00245 00527 00227


1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 014691198602 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 007491198603 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 012381198611 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 009801198612 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 007661198613 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 002851198621 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 006871198622 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 008381198623 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 010311198631 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 009061198632 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 006661198633 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386




























maintenance (C2)






Acknowledgments


References





















































1 Introduction













2 Literature Review












Potholeimages


Vehicle stationary


optics

Center





Center

RSE

company

OBU


Pothole information

Potholeinformation


or WAVE



severity)

or

















Input image

Binarization by HST

Segmented images



Candidaterefinement




Candidate region




119862V

=

1 if 119878 (1198721015840119888) gt 119879119904 Com (1198721015840


0 otherwise

(1)







com (1198721015840119888) =1198972

4120587119860 (2)



1198621015840

V =










OHI (ℎ119888 ℎ119887) =

119899

sum

119894=0

min (oh119894119888 oh119894119887) (4)









119901

=


1198881015840 ℎ119887) gt 119879119900 OHI (ℎ1015840

1198881015840 ℎ1015840



(5)




1198881015840 ℎ1015840





proposed method





ℎ(threshold for





119879ℎ


image119879std1015840 10

119879119904 512 119879ave 00119879com 005 119879

119900087

119879std 8 1198791199001015840 085





















(a) Asphalt images

(b) Concrete images







6 Conclusions





(a) Original images

(b) Close-up images



(a) Bright images

(b) Dark images








Acknowledgment


References


































Contents















1 Introduction


















References
















1 Introduction













1198881 1198882 119888


1 1198882 119888

119899are


1 V2 V

119896 in



Custo

mer

-veh

icle

assig

nmen

t

Opt

imiz

ed as

signm

ent

CV

c1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

CV

c1c2

cn

12

k

middot



Fitness = min119899

sum

119894=0

119899

sum

119895=0

119896

sum

V=1119889119894119895119883

V119894119895+ 1198891198990119883

V1198990

119894 = 119895 (1)

119899

sum

119894=0

119899

sum

119895=0

119883

V119894119895119903119894le 119876V 119894 = 119895 V isin 119881 (2)

119883

V119894119895

=


0 otherwise

(3)


119894119895is the









1 1198882 119888

119899 to










119894= 119883

1198941 119883

119894119871





119894119871 whereas


119871


119881

new119894119895


times (119866119895minus 119883119894119895)

119883

new119894119895= 119883119894119895+ 119881

new119894119895

(4)



1

Depot

310

8

2

95

7

6

4

Opt

imiz

ed sc

hedu

le

Opt

imiz

ed as

signm

ent

1

Depot

72

8

10

95

3

6

4

7

Depot

310

8

5

92

1

6

4

CV

c1c2

cn

12

k

middot




119883

119881

119894119883

119881

1198941119883

119881


119881

119894119899119883

119881

119894119899+1119883

119881

119894119899+2sdot sdot sdot 119883

119881

119894119899+119896minus1











119894= 119883

119881

1198941 119883

119881

119894119899 119883

119881

119894119899+119896minus1


1Cus2 Cus

119899Veh1Veh2 Veh

119896minus1) For

















119863

1 119883

119863

2) is shown



Key2 13 08 24 19 02 12 21

02 08 12 13 19 2 21 24Key


Vehicle assignment

Visit order

Veh 1

Veh1

Veh1 Veh2

Veh2

Cus1

Cus1

Cus1

Veh 2

Cus2

Cus2

Cus2

Veh 3

Cus3

Cus3

Cus3

Cus4

Cus4

Cus4

Cus5

Cus5

Cus5

Cus6

Cus6

Cus6

XiV

XiV



X119863 119883

119863

1119883

119863

2






119862 119910119862)) can be yielded by

119909119862=

sum

119896

119894=1119909

V119894+ 119909119874

119896 + 1

119910119862=

sum

119896

119894=1119910

V119894+ 119910119874

119896 + 1

(5)

F

O

DE

G

HA

I

C

J

B

K

Route-1

Route-2

Route-3


F

O

DE

G

HA

I

C

J

B

K

dE

dF

dD

dC

dB

dA

CC


In (5) 119909119862and 119910





119874are the




F

O

DE

B

C

JK

G

H

I

A

(a) 1198891 = 119874119861 + 119861119866minus 119874119866

F

O

DE

B

C

JK

G

H

I

A

(b) 1198892 = 119866119861 + 119861119867minus 119866119867

F

O

DE

C

J

A

K

G

H

IB

(c) 1198893 = 119867119861 + 119861119868 minus 119867119868

F

O

DE

B

C

J

A

K

G

H

I

(d) 1198894 = 119868119861 + 119861119874minus 119868119874








(3) Outer layer PSO



layer PSO





Start




End

Yes Yes

NoNo

YesNo


Initialize VVX

V

Update VVX

V

Initialize VDX

D

Update VDX

D

search(XV)

Fitness(X ) lt

Fitness(XV)

Update(SA)

Fitness( )

Updateand

Pass XD

to inner layer PSO

Fitness(XD) =

Fitness( )= XLSV

GVbest

XVnew

PVbest

XVnew X

Vnew

Updateand

GVbest

PVbest

GVbest

LSV

XVLS = local




119899









A-n54-k9 53 + 53 53 + 8

A-n64-k8 63 + 63 63 + 7







1= 2 and 119888

2= 1 and the


1= 1198882= 2 and 100



BKStimes 100 (6)


Avg Dev () =sum

119899

119894=1DEV119894

119899

(7)




1) RBI (LS

2)




0102030405060708090

Aver

age d

evia

tion

()

A-n3

2-k5

A-n3

3-k5

A-n3

6-k5

A-n4

5-k6

A-n4

5-k7

A-n5

5-k9

A-n6

0-k9

A-n6

2-k8

A-n6

4-k9

Aver

age

wo LSLS1

LS2LS3




Imp() =Fitness


Fitness119908119900

times 100 (8)





Fitn

ess

950

900

850

800

750

700

Iterations

0

50

100

150

200

250

300

350

400

450

500

550

600

650

700

750

800

850

900

950

1000

(a)

Fitn

ess

830

810

790

770

750

730

710

690

670

650

Iterations

0 5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

(b)


(a) (b)






(a) (b)



5 Conclusions






Acknowledgment


References




































1 Introduction



























HMM correction

Vehicle V1

Vehicle V2

Vehicle Vn



based on HMM

Entireroutes

Routerecommendation


Vehicle Vi

A set ofOutput

Input

RS(Tk Roadi)

RouteT119896



upcomingroutes


Rm1Rm2

Rm3

Rm4

Rm5

Rm1x

Rm2y

Rm3x Rm4x

Rm5y

Y

X0


Ri

Rj

Rk

Rix

Rjy

Rkx

Rky

Y

X0



119860119899

997888rarr 119861 = 1198771119909 (1198771119910)


(1)





= (1 3) (1 4) (3 4) (3 1)

(2)






Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd


Sichuan RdJinmao St

Longpan Rd

Jianning Rd

0 1 2 3 4 5 6 7 80

1

2

3

4

5

6

Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd

Central Rd

Xufu Alley

Sichuan Rd

Jinmao St

Longpan Rd

Jianning Rd

X

Y












1199051 lt (1 3) (1 4) (3 4) (3 1) gt

1199052 lt (3 1) (3 4) (1 4) (1 3) gt

1199053 lt (0 3) (1 3) (1 5) (4 5) gt

1199054 lt (0 3) (0 0) (0 4) (4 1) gt

1199055 lt (2 0) (2 1) (3 1) (3 2) (4 2) gt

1199051 lt (1 3) (1 4) (3 4) (3 1) gt

(3)














997888997888997888997888rarr119860119895119861119895 = (119909119895119899minus1199091198951 119910119895119899minus



cos ⟨ 119886119894 119886119895⟩ =

119886119894 sdot 119886119895

1003816100381610038161003816 1198861198941003816100381610038161003816 sdot10038161003816100381610038161003816119886119895

10038161003816100381610038161003816

= ((119909119894119899 minus 1199091198941) sdot (119910119894119899 minus 1199101198941)

+ (119909119895119899 minus 1199091198951) sdot (119910119895119899 minus 1199101198951))


2

sdotradic(119909119895119899 minus 1199091198951)2

+ (119910119895119899 minus 1199101198951)2

)

minus1

(4)


1





120587119894 =

Count (119904119863119894

) + 120582Count (119904New119863119894

)

sum119899

119895=1[Count (119904119863

119895

) + 120582Count (119904New119863119895

)]

(5)




1198961= 1199001198962)






119894

)

and Count(119904New119863119894



119875 (119904119894 | 119904119894minus1)

=

Count (119904119863119894minus1

119904119863119894

) + 120582Count (119904New119863119894minus1

119904New119863119894

) + 1

Count (119904119863119894minus1

) + 120582Count (119904New119863119894minus1

) + 119898

(6)


119904119863119894


119904New119863119894

)



119875 (V119895 | 119904119894)

=

Count (119904119863119894minus1

V119863119894

) + 120582Count (119904New119863119894minus1

VNew119863119894

) + 1

Count (119904119863119894

) + 120582Count (119904New119863119894

) + 119899

(7)


V119863119894


VNew119863119894

)






sequence

Viterbialgorithm










119901119894(1199091) 119901119894(119909

2) 119901119894(119909



1) 119901119894(119910

2) 119901119894(119910

119899) (1198632 notin 119881) and the











(b) Suppose that 1198632 = 119901119894(1199101) 119901119894(119910

2) 119901119894119896 then


(c) Suppose that 1198632 = 1199011198941 119901119894(1199102) 119901119894(119910

119899) then



1) 119901119894(119910

2) 119901119894(119910

119899) and 1199101 = 1 119910119899


1)

119901119894(1199102) 119901119894(119910





















119875119894 =sum119899

119896=1119863(119877119896 119862119877)

sum119899

119905=1Dist 1003816100381610038161003816119877119905

1003816100381610038161003816

times 100 (8)



1198751 =Repeat (1198771 1198771) + Repeat (1198773 1198771)

Dist 100381610038161003816100381611987711003816100381610038161003816 + Dist 10038161003816100381610038161198773

1003816100381610038161003816

times 100

=6 + 1

6 + 7times 100 = 5385

(9)












(a) (b)


100908070605040302010

01009080706050403020100

Trip completed ()

Cor

rect

pre

dict

ion

()


Cor

rect

pre

dict

ion

()

100908070605040302010

01009080706050403020100

Trip completed ()


Cor

rect

pre

dict

ion

()


100908070605040302010

0


120582 = 05

120582 = 02


Cor

rect

pre

dict

ion

()

100908070605040302010

0



120582 = 05

120582 = 02







90

80

70

60

50

40

30

20

10

0Other time periods

Cor

rect

pre

dict

ion

()

Time of day







7 Conclusion






Acknowledgments


References

























1 Introduction



















1198811198732

1198811198733

1198811198734

1198811198731

(1198811198731 1198811198731) (119881119873

1 1198811198732) (119881119873

1 1198811198733) (119881119873

1 1198811198734)

1198811198732

(1198811198732 1198811198731) (119881119873

2 1198811198732) (119881119873

2 1198811198733) (119881119873

2 1198811198734)

1198811198733

(1198811198733 1198811198731) (119881119873

3 1198811198732) (119881119873

3 1198811198733) (119881119873

3 1198811198734)

1198811198734

(1198811198734 1198811198731) (119881119873

4 1198811198732) (119881119873

4 1198811198733) (119881119873

4 1198811198734)



1 1199051⟩ rarr ⟨119873


119899 119905119899⟩)


1rarr 119873


119899)




119899) where

tr(119896+1)



tr(119896+1)


119896




1rarr 119873


119873119896minus1


1








Subregion 1

Subregion 2

Subregion 3

Subregion 4

Node 1Node 4

Node 2Node 3

Virtual node 4


Virtual node 1

Virtual node 4


Virtual node 1

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Physical topology

Logical topology


Link 2

Link 5

Link 1

Path 1 Path 2

Link 3

Link 4

Path 3 Path 4


Link 1


Link 5

Path 1

Path 2

Path 3

Path 4

Path 1Link 1

Link 3

Link 4

Path 2


Path 3Link 2

Link 3

Link 4 Path 2

Link 3Link 2


Path 1 Path 3

Path 4

Link 4

Path 2

(b) Index






[[[[[[[[

[

11990911

11990912


1199091119908

11990921

11990922


1199092119908


119909119879minus11

119909119879minus12


119909119879minus1119908

1199091198791

1199091198792


119909119879119908

]]]]]]]]

]

(1)








119885119871 119884119898

(119898 = 1 119871) (2)


119871 119884119898(119898 = 1 119871) in every scale is





119871 119898(119898 = 1 119871) at all scales



SPE119894=

119882

sum

119895=1

(119909119894119895minus 119909119894119895)2

(3)






1205852

120572= 1206011

[[

[

119888120572radic21206012ℎ2

0

1206011

+ 1 +1206012ℎ0(ℎ0minus 1)

1206012

1

]]

]

1ℎ0

(4)


1120601331206012

2 120601119894= sum119882

119895=119903+1120582119894

119895 119894 = 1 2 3 120582

119895is


120572is the 1 minus 120572





Transform

Physical topology

Logical topology

Taxi GPSdata

Traffic flowdata


transformPCA




Judge

Anomalousposition



119905

Thus

119876119905+1= 120573119876119905+ (1 minus 120573)119876

119905 (5)




120595 = 120583119904+ 119871 times 120590

119904radic

120573

(2 minus 120573) 119879 (6)
























119889119862119903 where 119862





GPS data



















1 432012 5927 6297 83172 632012 6418 6452 75863 732012 5344 7020 8849

32 1152012 6374 7270 8237

74 3182012 4728 7737 7888Average 119877 value 5633 6329 7462

050

100150200250300350400450500

Traffi

c flow


t

700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

20

40

60

80

100

120

140

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

50

100

150

200

250

300

350

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

10

20

30

40

50

60

70

80

()


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

845

ndash900

900

ndash915

915

ndash930











0

2000

4000

6000

8000

10000

12000

14000

16000


Proc

essin

g tim

e (m

s)



4 Conclusions












Acknowledgments



References











































1 Introduction










2 Literature Review









Organization


[35ndash43]

Project base


[36 37 40 41 43]

Systemtechnology


[35ndash43]

Consultants


[35ndash37 39]

Externalenvironment


[38 42]



3 Methodology








Organization





Project base





Systemtechnology






Consultant





Externalenvironment









Normalized Z(X)




each factor







DEMATEL

ANP





Z =

[[[[[[[

[

1199111111991112sdot sdot sdot 119911

1119899

1199112111991122sdot sdot sdot 119911

2119899

11991111989911199111198992sdot sdot sdot 119911

119899119899

]]]]]]]

]

(1)


120582 =

1

max1le119894le119899sum119899

119895=1119885119894119895

(119894 119895 = 1 2 119899)

X = 120582 sdot Z(2)


lim119870rarrinfin

X119870 = 0


(3)



Z =119860

119861

119862

119863

((

(

119860

0

3

3

3

119861

2

0

1

2

119862

2

2

0

2

119863

2

1

2

0

))

)

(4)


X =119860

119861

119862

119863

((

(

119860

0000

0337

0326

0337

119861

0233

0000

0116

0198

119862

0279

0198

0000

0198

119863

0233

0116

0244

0000

))

)

T =

119860

119861

119862

119863

(

119860

0628

0817

0839

0876

119861

0580

0356

0483

0559

119862

0691

0593

0449

0637

119863

0615

0493

0605

0424

)

119889

2513

2259

2377

2497

119903 3159 1979 2370 2137

(5)





Table 3


119860 2513 3159 5673 1 minus06462119861 2259 1979 4238 4 02796119862 2377 2370 4746 2 00068119863 2496 2137 4633 3 03598

Table 4

119860 119861 119862 119863

119860 0199 0293 0291 0288119861 0259 0180 0250 0231119862 0266 0244 0190 0283119863 0277 0283 0269 0199





Table 5


Overallrankings


A B C DA 0628 0580 0691 0615B 0817 0256 0593 0493C 0839 0483 0449 0605D 0876 0559 0637 0424

T =

Figure 2

DA

C

Figure 3

4 Empirical Study



























43 Result Analysis






Aspects 119860 119861 119862 119863

119860 00000 20000 24000 20000119861 29000 00000 17000 10000119862 28000 10000 00000 21000119863 29000 17000 17000 00000


Aspects 119860 119861 119862 119863

119860 00000 02326 02791 02326119861 03372 00000 01977 01163119862 03256 01163 00000 02442119863 03372 01977 01977 00000


Aspects 119860 119861 119862 119863 119889

119860 06278 05803 06905 06146 25132119861 08166 03563 05933 04925 22587119862 08389 04832 04492 06052 23765119863 08761 05593 06366 04242 24963119903 31593 19791 23697 21365


Aspects 119889 119903 119889 + 119903 119889 minus 119903

119860 25132 31593 56725 minus06462119861 22587 19791 42378 02796119862 23765 23697 47461 00068119863 24963 21365 46328 03598


Aspects 119860 119861 119862 119863

119860 01987 02932 02914 02877119861 02585 01800 02504 02305119862 02655 02442 01896 02832119863 02773 02826 02686 01986


Aspects 119860 119861 119862 119863

119860 02662 02662 02662 02662119861 02312 02312 02312 02312119862 02464 02464 02464 02464119863 02562 02562 02562 02562




Overallrankings



Organizations(A)


(D)System

technology (C)

Project base (B)









Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 40000 40000 40000 24000 20000 28000 40000 20000 40000 30000 400001198602 30000 00000 20000 18000 22000 20000 30000 00000 00000 00000 30000 200001198603 39000 20000 00000 30000 19000 21000 24000 25000 25000 36000 20000 220001198611 16000 27000 30000 00000 19000 30000 23000 20000 10000 17000 40000 290001198612 10000 16000 10000 10000 00000 30000 24000 10000 20000 24000 26000 180001198613 01000 15000 12000 02000 00000 00000 21000 00000 01000 04000 10000 140001198621 20000 18000 20000 14000 16000 10000 00000 30000 00000 00000 10000 300001198622 10000 10000 25000 14000 18000 19000 27000 00000 20000 25000 15000 140001198623 25000 20000 29000 20000 19000 20000 26000 30000 00000 29000 10000 200001198631 30000 30000 30000 08000 23000 30000 24000 00000 00000 00000 40000 300001198632 29000 20000 00000 06000 16000 26000 21000 09000 00000 31000 00000 130001198633 18000 13000 14000 02000 09000 03000 10000 00000 00000 00000 18000 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 01105 01105 01105 00663 00552 00773 01105 00552 01105 00829 011051198602 00829 00000 00552 00497 00608 00552 00829 00000 00000 00000 00829 005521198603 01077 00552 00000 00829 00525 00580 00663 00691 00691 00994 00552 006081198611 00442 00746 00829 00000 00525 00829 00635 00552 00276 00470 01105 008011198612 00276 00442 00276 00276 00000 00829 00663 00276 00552 00663 00718 004971198613 00028 00414 00331 00055 00000 00000 00580 00000 00028 00110 00276 003871198621 00552 00497 00552 00387 00442 00276 00000 00829 00000 00000 00276 008291198622 00276 00276 00691 00387 00497 00525 00746 00000 00552 00691 00414 003871198623 00691 00552 00801 00552 00525 00552 00718 00829 00000 00801 00276 005521198631 00829 00829 00829 00221 00635 00829 00663 00000 00000 00000 01105 008291198632 00801 00552 00000 00166 00442 00718 00580 00249 00000 00856 00000 003591198633 00497 00359 00387 00055 00249 00083 00276 00000 00000 00000 00497 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633 119889

1198601 01250 02233 02211 01894 01618 01718 02066 01854 01023 02070 02120 02347 224041198602 01424 00664 01129 00954 01090 01150 01484 00500 00274 00582 01475 01249 119751198603 01991 01544 01007 01508 01311 01526 01722 01371 01064 01808 01621 01682 181551198611 01294 01542 01563 00593 01173 01606 01537 01094 00602 01181 01938 01663 157861198612 00915 01064 00878 00699 00504 01407 01334 00697 00753 01158 01356 01170 119361198613 00316 00647 00553 00240 00212 00230 00828 00183 00112 00296 00533 00655 048041198621 01085 01029 01082 00795 00883 00807 00629 01188 00273 00512 00885 01398 105671198622 00962 00947 01311 00855 01019 01164 01447 00487 00806 01242 01120 01116 124771198623 01521 01393 01621 01165 01205 01368 01635 01403 00376 01511 01215 01482 158951198631 01614 01602 01518 00802 01243 01561 01513 00561 00320 00695 01910 01665 150021198632 01319 01132 00593 00575 00890 01249 01196 00625 00217 01277 00654 01007 107341198633 00816 00679 00671 00315 00508 00399 00624 00252 00143 00309 00824 00359 05899119903 14507 14476 14136 10395 11656 14185 16015 10217 05964 12641 15651 15790




3

Impo

rtan

ce ra

nkin

g

Noncritical

Critical1

7

8

12


















Criteria 119889 119903 119889 + 119903 119889 minus 119903




5 Conclusions









1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00862 01542 01564 01822 01388 01211 01290 01815 01715 01637 01355 014861198602 00982 00459 00799 00917 00935 00810 00927 00490 00459 00461 00943 007911198603 01372 01066 00712 01451 01125 01076 01075 01342 01784 01430 01036 010651198611 00892 01065 01105 00570 01007 01132 00960 01071 01009 00934 01238 010531198612 00631 00735 00621 00673 00432 00992 00833 00682 01263 00916 00866 007411198613 00218 00447 00391 00230 00182 00162 00517 00179 00188 00234 00341 004151198621 00748 00711 00765 00765 00757 00569 00393 01163 00458 00405 00566 008851198622 00663 00654 00927 00822 00874 00821 00904 00477 01352 00983 00716 007071198623 01048 00963 01147 01121 01034 00965 01021 01374 00630 01195 00776 009381198631 01112 01106 01074 00771 01066 01101 00945 00549 00537 00549 01220 010541198632 00909 00782 00420 00554 00764 00880 00747 00612 00364 01011 00418 006381198633 00562 00469 00474 00303 00436 00281 00390 00247 00240 00245 00527 00227


1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 014691198602 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 007491198603 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 012381198611 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 009801198612 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 007661198613 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 002851198621 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 006871198622 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 008381198623 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 010311198631 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 009061198632 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 006661198633 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386




























maintenance (C2)






Acknowledgments


References





















































1 Introduction













2 Literature Review












Potholeimages


Vehicle stationary


optics

Center





Center

RSE

company

OBU


Pothole information

Potholeinformation


or WAVE



severity)

or

















Input image

Binarization by HST

Segmented images



Candidaterefinement




Candidate region




119862V

=

1 if 119878 (1198721015840119888) gt 119879119904 Com (1198721015840


0 otherwise

(1)







com (1198721015840119888) =1198972

4120587119860 (2)



1198621015840

V =










OHI (ℎ119888 ℎ119887) =

119899

sum

119894=0

min (oh119894119888 oh119894119887) (4)









119901

=


1198881015840 ℎ119887) gt 119879119900 OHI (ℎ1015840

1198881015840 ℎ1015840



(5)




1198881015840 ℎ1015840





proposed method





ℎ(threshold for





119879ℎ


image119879std1015840 10

119879119904 512 119879ave 00119879com 005 119879

119900087

119879std 8 1198791199001015840 085





















(a) Asphalt images

(b) Concrete images







6 Conclusions





(a) Original images

(b) Close-up images



(a) Bright images

(b) Dark images








Acknowledgment


References































Contents















1 Introduction


















References
















1 Introduction













1198881 1198882 119888


1 1198882 119888

119899are


1 V2 V

119896 in



Custo

mer

-veh

icle

assig

nmen

t

Opt

imiz

ed as

signm

ent

CV

c1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

C Vc1c2

cn

12

k

middot

CV

c1c2

cn

12

k

middot



Fitness = min119899

sum

119894=0

119899

sum

119895=0

119896

sum

V=1119889119894119895119883

V119894119895+ 1198891198990119883

V1198990

119894 = 119895 (1)

119899

sum

119894=0

119899

sum

119895=0

119883

V119894119895119903119894le 119876V 119894 = 119895 V isin 119881 (2)

119883

V119894119895

=


0 otherwise

(3)


119894119895is the









1 1198882 119888

119899 to










119894= 119883

1198941 119883

119894119871





119894119871 whereas


119871


119881

new119894119895


times (119866119895minus 119883119894119895)

119883

new119894119895= 119883119894119895+ 119881

new119894119895

(4)



1

Depot

310

8

2

95

7

6

4

Opt

imiz

ed sc

hedu

le

Opt

imiz

ed as

signm

ent

1

Depot

72

8

10

95

3

6

4

7

Depot

310

8

5

92

1

6

4

CV

c1c2

cn

12

k

middot




119883

119881

119894119883

119881

1198941119883

119881


119881

119894119899119883

119881

119894119899+1119883

119881

119894119899+2sdot sdot sdot 119883

119881

119894119899+119896minus1











119894= 119883

119881

1198941 119883

119881

119894119899 119883

119881

119894119899+119896minus1


1Cus2 Cus

119899Veh1Veh2 Veh

119896minus1) For

















119863

1 119883

119863

2) is shown



Key2 13 08 24 19 02 12 21

02 08 12 13 19 2 21 24Key


Vehicle assignment

Visit order

Veh 1

Veh1

Veh1 Veh2

Veh2

Cus1

Cus1

Cus1

Veh 2

Cus2

Cus2

Cus2

Veh 3

Cus3

Cus3

Cus3

Cus4

Cus4

Cus4

Cus5

Cus5

Cus5

Cus6

Cus6

Cus6

XiV

XiV



X119863 119883

119863

1119883

119863

2






119862 119910119862)) can be yielded by

119909119862=

sum

119896

119894=1119909

V119894+ 119909119874

119896 + 1

119910119862=

sum

119896

119894=1119910

V119894+ 119910119874

119896 + 1

(5)

F

O

DE

G

HA

I

C

J

B

K

Route-1

Route-2

Route-3


F

O

DE

G

HA

I

C

J

B

K

dE

dF

dD

dC

dB

dA

CC


In (5) 119909119862and 119910





119874are the




F

O

DE

B

C

JK

G

H

I

A

(a) 1198891 = 119874119861 + 119861119866minus 119874119866

F

O

DE

B

C

JK

G

H

I

A

(b) 1198892 = 119866119861 + 119861119867minus 119866119867

F

O

DE

C

J

A

K

G

H

IB

(c) 1198893 = 119867119861 + 119861119868 minus 119867119868

F

O

DE

B

C

J

A

K

G

H

I

(d) 1198894 = 119868119861 + 119861119874minus 119868119874








(3) Outer layer PSO



layer PSO





Start




End

Yes Yes

NoNo

YesNo


Initialize VVX

V

Update VVX

V

Initialize VDX

D

Update VDX

D

search(XV)

Fitness(X ) lt

Fitness(XV)

Update(SA)

Fitness( )

Updateand

Pass XD

to inner layer PSO

Fitness(XD) =

Fitness( )= XLSV

GVbest

XVnew

PVbest

XVnew X

Vnew

Updateand

GVbest

PVbest

GVbest

LSV

XVLS = local




119899









A-n54-k9 53 + 53 53 + 8

A-n64-k8 63 + 63 63 + 7







1= 2 and 119888

2= 1 and the


1= 1198882= 2 and 100



BKStimes 100 (6)


Avg Dev () =sum

119899

119894=1DEV119894

119899

(7)




1) RBI (LS

2)




0102030405060708090

Aver

age d

evia

tion

()

A-n3

2-k5

A-n3

3-k5

A-n3

6-k5

A-n4

5-k6

A-n4

5-k7

A-n5

5-k9

A-n6

0-k9

A-n6

2-k8

A-n6

4-k9

Aver

age

wo LSLS1

LS2LS3




Imp() =Fitness


Fitness119908119900

times 100 (8)





Fitn

ess

950

900

850

800

750

700

Iterations

0

50

100

150

200

250

300

350

400

450

500

550

600

650

700

750

800

850

900

950

1000

(a)

Fitn

ess

830

810

790

770

750

730

710

690

670

650

Iterations

0 5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

(b)


(a) (b)






(a) (b)



5 Conclusions






Acknowledgment


References




































1 Introduction



























HMM correction

Vehicle V1

Vehicle V2

Vehicle Vn



based on HMM

Entireroutes

Routerecommendation


Vehicle Vi

A set ofOutput

Input

RS(Tk Roadi)

RouteT119896



upcomingroutes


Rm1Rm2

Rm3

Rm4

Rm5

Rm1x

Rm2y

Rm3x Rm4x

Rm5y

Y

X0


Ri

Rj

Rk

Rix

Rjy

Rkx

Rky

Y

X0



119860119899

997888rarr 119861 = 1198771119909 (1198771119910)


(1)





= (1 3) (1 4) (3 4) (3 1)

(2)






Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd


Sichuan RdJinmao St

Longpan Rd

Jianning Rd

0 1 2 3 4 5 6 7 80

1

2

3

4

5

6

Fujian Rd

Zhongfu Rd

Heilongjiang Rd

Zhongshan North Rd

Nanrui Rd

New Mofan Rd

Central Rd

Xufu Alley

Sichuan Rd

Jinmao St

Longpan Rd

Jianning Rd

X

Y












1199051 lt (1 3) (1 4) (3 4) (3 1) gt

1199052 lt (3 1) (3 4) (1 4) (1 3) gt

1199053 lt (0 3) (1 3) (1 5) (4 5) gt

1199054 lt (0 3) (0 0) (0 4) (4 1) gt

1199055 lt (2 0) (2 1) (3 1) (3 2) (4 2) gt

1199051 lt (1 3) (1 4) (3 4) (3 1) gt

(3)














997888997888997888997888rarr119860119895119861119895 = (119909119895119899minus1199091198951 119910119895119899minus



cos ⟨ 119886119894 119886119895⟩ =

119886119894 sdot 119886119895

1003816100381610038161003816 1198861198941003816100381610038161003816 sdot10038161003816100381610038161003816119886119895

10038161003816100381610038161003816

= ((119909119894119899 minus 1199091198941) sdot (119910119894119899 minus 1199101198941)

+ (119909119895119899 minus 1199091198951) sdot (119910119895119899 minus 1199101198951))


2

sdotradic(119909119895119899 minus 1199091198951)2

+ (119910119895119899 minus 1199101198951)2

)

minus1

(4)


1





120587119894 =

Count (119904119863119894

) + 120582Count (119904New119863119894

)

sum119899

119895=1[Count (119904119863

119895

) + 120582Count (119904New119863119895

)]

(5)




1198961= 1199001198962)






119894

)

and Count(119904New119863119894



119875 (119904119894 | 119904119894minus1)

=

Count (119904119863119894minus1

119904119863119894

) + 120582Count (119904New119863119894minus1

119904New119863119894

) + 1

Count (119904119863119894minus1

) + 120582Count (119904New119863119894minus1

) + 119898

(6)


119904119863119894


119904New119863119894

)



119875 (V119895 | 119904119894)

=

Count (119904119863119894minus1

V119863119894

) + 120582Count (119904New119863119894minus1

VNew119863119894

) + 1

Count (119904119863119894

) + 120582Count (119904New119863119894

) + 119899

(7)


V119863119894


VNew119863119894

)






sequence

Viterbialgorithm










119901119894(1199091) 119901119894(119909

2) 119901119894(119909



1) 119901119894(119910

2) 119901119894(119910

119899) (1198632 notin 119881) and the











(b) Suppose that 1198632 = 119901119894(1199101) 119901119894(119910

2) 119901119894119896 then


(c) Suppose that 1198632 = 1199011198941 119901119894(1199102) 119901119894(119910

119899) then



1) 119901119894(119910

2) 119901119894(119910

119899) and 1199101 = 1 119910119899


1)

119901119894(1199102) 119901119894(119910





















119875119894 =sum119899

119896=1119863(119877119896 119862119877)

sum119899

119905=1Dist 1003816100381610038161003816119877119905

1003816100381610038161003816

times 100 (8)



1198751 =Repeat (1198771 1198771) + Repeat (1198773 1198771)

Dist 100381610038161003816100381611987711003816100381610038161003816 + Dist 10038161003816100381610038161198773

1003816100381610038161003816

times 100

=6 + 1

6 + 7times 100 = 5385

(9)












(a) (b)


100908070605040302010

01009080706050403020100

Trip completed ()

Cor

rect

pre

dict

ion

()


Cor

rect

pre

dict

ion

()

100908070605040302010

01009080706050403020100

Trip completed ()


Cor

rect

pre

dict

ion

()


100908070605040302010

0


120582 = 05

120582 = 02


Cor

rect

pre

dict

ion

()

100908070605040302010

0



120582 = 05

120582 = 02







90

80

70

60

50

40

30

20

10

0Other time periods

Cor

rect

pre

dict

ion

()

Time of day







7 Conclusion






Acknowledgments


References

























1 Introduction



















1198811198732

1198811198733

1198811198734

1198811198731

(1198811198731 1198811198731) (119881119873

1 1198811198732) (119881119873

1 1198811198733) (119881119873

1 1198811198734)

1198811198732

(1198811198732 1198811198731) (119881119873

2 1198811198732) (119881119873

2 1198811198733) (119881119873

2 1198811198734)

1198811198733

(1198811198733 1198811198731) (119881119873

3 1198811198732) (119881119873

3 1198811198733) (119881119873

3 1198811198734)

1198811198734

(1198811198734 1198811198731) (119881119873

4 1198811198732) (119881119873

4 1198811198733) (119881119873

4 1198811198734)



1 1199051⟩ rarr ⟨119873


119899 119905119899⟩)


1rarr 119873


119899)




119899) where

tr(119896+1)



tr(119896+1)


119896




1rarr 119873


119873119896minus1


1








Subregion 1

Subregion 2

Subregion 3

Subregion 4

Node 1Node 4

Node 2Node 3

Virtual node 4


Virtual node 1

Virtual node 4


Virtual node 1

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Virtual node 1

Virtual node 4

Virtual node 2

Virtual node 3

Physical topology

Logical topology


Link 2

Link 5

Link 1

Path 1 Path 2

Link 3

Link 4

Path 3 Path 4


Link 1


Link 5

Path 1

Path 2

Path 3

Path 4

Path 1Link 1

Link 3

Link 4

Path 2


Path 3Link 2

Link 3

Link 4 Path 2

Link 3Link 2


Path 1 Path 3

Path 4

Link 4

Path 2

(b) Index






[[[[[[[[

[

11990911

11990912


1199091119908

11990921

11990922


1199092119908


119909119879minus11

119909119879minus12


119909119879minus1119908

1199091198791

1199091198792


119909119879119908

]]]]]]]]

]

(1)








119885119871 119884119898

(119898 = 1 119871) (2)


119871 119884119898(119898 = 1 119871) in every scale is





119871 119898(119898 = 1 119871) at all scales



SPE119894=

119882

sum

119895=1

(119909119894119895minus 119909119894119895)2

(3)






1205852

120572= 1206011

[[

[

119888120572radic21206012ℎ2

0

1206011

+ 1 +1206012ℎ0(ℎ0minus 1)

1206012

1

]]

]

1ℎ0

(4)


1120601331206012

2 120601119894= sum119882

119895=119903+1120582119894

119895 119894 = 1 2 3 120582

119895is


120572is the 1 minus 120572





Transform

Physical topology

Logical topology

Taxi GPSdata

Traffic flowdata


transformPCA




Judge

Anomalousposition



119905

Thus

119876119905+1= 120573119876119905+ (1 minus 120573)119876

119905 (5)




120595 = 120583119904+ 119871 times 120590

119904radic

120573

(2 minus 120573) 119879 (6)
























119889119862119903 where 119862





GPS data



















1 432012 5927 6297 83172 632012 6418 6452 75863 732012 5344 7020 8849

32 1152012 6374 7270 8237

74 3182012 4728 7737 7888Average 119877 value 5633 6329 7462

050

100150200250300350400450500

Traffi

c flow


t

700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

20

40

60

80

100

120

140

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

50

100

150

200

250

300

350

Traffi

c flow


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

900

ndash915

915

ndash930

845

ndash900


t

0

10

20

30

40

50

60

70

80

()


700

ndash715

715

ndash730

730

ndash745

745

ndash800

800

ndash815

815

ndash830

830

ndash845

845

ndash900

900

ndash915

915

ndash930











0

2000

4000

6000

8000

10000

12000

14000

16000


Proc

essin

g tim

e (m

s)



4 Conclusions












Acknowledgments



References











































1 Introduction










2 Literature Review









Organization


[35ndash43]

Project base


[36 37 40 41 43]

Systemtechnology


[35ndash43]

Consultants


[35ndash37 39]

Externalenvironment


[38 42]



3 Methodology








Organization





Project base





Systemtechnology






Consultant





Externalenvironment









Normalized Z(X)




each factor







DEMATEL

ANP





Z =

[[[[[[[

[

1199111111991112sdot sdot sdot 119911

1119899

1199112111991122sdot sdot sdot 119911

2119899

11991111989911199111198992sdot sdot sdot 119911

119899119899

]]]]]]]

]

(1)


120582 =

1

max1le119894le119899sum119899

119895=1119885119894119895

(119894 119895 = 1 2 119899)

X = 120582 sdot Z(2)


lim119870rarrinfin

X119870 = 0


(3)



Z =119860

119861

119862

119863

((

(

119860

0

3

3

3

119861

2

0

1

2

119862

2

2

0

2

119863

2

1

2

0

))

)

(4)


X =119860

119861

119862

119863

((

(

119860

0000

0337

0326

0337

119861

0233

0000

0116

0198

119862

0279

0198

0000

0198

119863

0233

0116

0244

0000

))

)

T =

119860

119861

119862

119863

(

119860

0628

0817

0839

0876

119861

0580

0356

0483

0559

119862

0691

0593

0449

0637

119863

0615

0493

0605

0424

)

119889

2513

2259

2377

2497

119903 3159 1979 2370 2137

(5)





Table 3


119860 2513 3159 5673 1 minus06462119861 2259 1979 4238 4 02796119862 2377 2370 4746 2 00068119863 2496 2137 4633 3 03598

Table 4

119860 119861 119862 119863

119860 0199 0293 0291 0288119861 0259 0180 0250 0231119862 0266 0244 0190 0283119863 0277 0283 0269 0199





Table 5


Overallrankings


A B C DA 0628 0580 0691 0615B 0817 0256 0593 0493C 0839 0483 0449 0605D 0876 0559 0637 0424

T =

Figure 2

DA

C

Figure 3

4 Empirical Study



























43 Result Analysis






Aspects 119860 119861 119862 119863

119860 00000 20000 24000 20000119861 29000 00000 17000 10000119862 28000 10000 00000 21000119863 29000 17000 17000 00000


Aspects 119860 119861 119862 119863

119860 00000 02326 02791 02326119861 03372 00000 01977 01163119862 03256 01163 00000 02442119863 03372 01977 01977 00000


Aspects 119860 119861 119862 119863 119889

119860 06278 05803 06905 06146 25132119861 08166 03563 05933 04925 22587119862 08389 04832 04492 06052 23765119863 08761 05593 06366 04242 24963119903 31593 19791 23697 21365


Aspects 119889 119903 119889 + 119903 119889 minus 119903

119860 25132 31593 56725 minus06462119861 22587 19791 42378 02796119862 23765 23697 47461 00068119863 24963 21365 46328 03598


Aspects 119860 119861 119862 119863

119860 01987 02932 02914 02877119861 02585 01800 02504 02305119862 02655 02442 01896 02832119863 02773 02826 02686 01986


Aspects 119860 119861 119862 119863

119860 02662 02662 02662 02662119861 02312 02312 02312 02312119862 02464 02464 02464 02464119863 02562 02562 02562 02562




Overallrankings



Organizations(A)


(D)System

technology (C)

Project base (B)









Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 40000 40000 40000 24000 20000 28000 40000 20000 40000 30000 400001198602 30000 00000 20000 18000 22000 20000 30000 00000 00000 00000 30000 200001198603 39000 20000 00000 30000 19000 21000 24000 25000 25000 36000 20000 220001198611 16000 27000 30000 00000 19000 30000 23000 20000 10000 17000 40000 290001198612 10000 16000 10000 10000 00000 30000 24000 10000 20000 24000 26000 180001198613 01000 15000 12000 02000 00000 00000 21000 00000 01000 04000 10000 140001198621 20000 18000 20000 14000 16000 10000 00000 30000 00000 00000 10000 300001198622 10000 10000 25000 14000 18000 19000 27000 00000 20000 25000 15000 140001198623 25000 20000 29000 20000 19000 20000 26000 30000 00000 29000 10000 200001198631 30000 30000 30000 08000 23000 30000 24000 00000 00000 00000 40000 300001198632 29000 20000 00000 06000 16000 26000 21000 09000 00000 31000 00000 130001198633 18000 13000 14000 02000 09000 03000 10000 00000 00000 00000 18000 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00000 01105 01105 01105 00663 00552 00773 01105 00552 01105 00829 011051198602 00829 00000 00552 00497 00608 00552 00829 00000 00000 00000 00829 005521198603 01077 00552 00000 00829 00525 00580 00663 00691 00691 00994 00552 006081198611 00442 00746 00829 00000 00525 00829 00635 00552 00276 00470 01105 008011198612 00276 00442 00276 00276 00000 00829 00663 00276 00552 00663 00718 004971198613 00028 00414 00331 00055 00000 00000 00580 00000 00028 00110 00276 003871198621 00552 00497 00552 00387 00442 00276 00000 00829 00000 00000 00276 008291198622 00276 00276 00691 00387 00497 00525 00746 00000 00552 00691 00414 003871198623 00691 00552 00801 00552 00525 00552 00718 00829 00000 00801 00276 005521198631 00829 00829 00829 00221 00635 00829 00663 00000 00000 00000 01105 008291198632 00801 00552 00000 00166 00442 00718 00580 00249 00000 00856 00000 003591198633 00497 00359 00387 00055 00249 00083 00276 00000 00000 00000 00497 00000


Criteria 1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633 119889

1198601 01250 02233 02211 01894 01618 01718 02066 01854 01023 02070 02120 02347 224041198602 01424 00664 01129 00954 01090 01150 01484 00500 00274 00582 01475 01249 119751198603 01991 01544 01007 01508 01311 01526 01722 01371 01064 01808 01621 01682 181551198611 01294 01542 01563 00593 01173 01606 01537 01094 00602 01181 01938 01663 157861198612 00915 01064 00878 00699 00504 01407 01334 00697 00753 01158 01356 01170 119361198613 00316 00647 00553 00240 00212 00230 00828 00183 00112 00296 00533 00655 048041198621 01085 01029 01082 00795 00883 00807 00629 01188 00273 00512 00885 01398 105671198622 00962 00947 01311 00855 01019 01164 01447 00487 00806 01242 01120 01116 124771198623 01521 01393 01621 01165 01205 01368 01635 01403 00376 01511 01215 01482 158951198631 01614 01602 01518 00802 01243 01561 01513 00561 00320 00695 01910 01665 150021198632 01319 01132 00593 00575 00890 01249 01196 00625 00217 01277 00654 01007 107341198633 00816 00679 00671 00315 00508 00399 00624 00252 00143 00309 00824 00359 05899119903 14507 14476 14136 10395 11656 14185 16015 10217 05964 12641 15651 15790




3

Impo

rtan

ce ra

nkin

g

Noncritical

Critical1

7

8

12


















Criteria 119889 119903 119889 + 119903 119889 minus 119903




5 Conclusions









1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 00862 01542 01564 01822 01388 01211 01290 01815 01715 01637 01355 014861198602 00982 00459 00799 00917 00935 00810 00927 00490 00459 00461 00943 007911198603 01372 01066 00712 01451 01125 01076 01075 01342 01784 01430 01036 010651198611 00892 01065 01105 00570 01007 01132 00960 01071 01009 00934 01238 010531198612 00631 00735 00621 00673 00432 00992 00833 00682 01263 00916 00866 007411198613 00218 00447 00391 00230 00182 00162 00517 00179 00188 00234 00341 004151198621 00748 00711 00765 00765 00757 00569 00393 01163 00458 00405 00566 008851198622 00663 00654 00927 00822 00874 00821 00904 00477 01352 00983 00716 007071198623 01048 00963 01147 01121 01034 00965 01021 01374 00630 01195 00776 009381198631 01112 01106 01074 00771 01066 01101 00945 00549 00537 00549 01220 010541198632 00909 00782 00420 00554 00764 00880 00747 00612 00364 01011 00418 006381198633 00562 00469 00474 00303 00436 00281 00390 00247 00240 00245 00527 00227


1198601 1198602 1198603 1198611 1198612 1198613 1198621 1198622 1198623 1198631 1198632 1198633

1198601 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 01469 014691198602 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 00749 007491198603 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 01238 012381198611 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 00980 009801198612 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 00766 007661198613 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 00285 002851198621 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 00687 006871198622 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 00838 008381198623 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 01031 010311198631 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 00906 009061198632 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 00666 006661198633 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386 00386




























maintenance (C2)






Acknowledgments


References





















































1 Introduction













2 Literature Review












Potholeimages


Vehicle stationary


optics

Center





Center

RSE

company

OBU


Pothole information

Potholeinformation


or WAVE



severity)

or

















Input image

Binarization by HST

Segmented images



Candidaterefinement




Candidate region




119862V

=

1 if 119878 (1198721015840119888) gt 119879119904 Com (1198721015840


0 otherwise

(1)







com (1198721015840119888) =1198972

4120587119860 (2)



1198621015840

V =










OHI (ℎ119888 ℎ119887) =

119899

sum

119894=0

min (oh119894119888 oh119894119887) (4)









119901

=


1198881015840 ℎ119887) gt 119879119900 OHI (ℎ1015840

1198881015840 ℎ1015840



(5)




1198881015840 ℎ1015840





proposed method





ℎ(threshold for





119879ℎ


image119879std1015840 10

119879119904 512 119879ave 00119879com 005 119879

119900087

119879std 8 1198791199001015840 085





















(a) Asphalt images

(b) Concrete images







6 Conclusions





(a) Original images

(b) Close-up images



(a) Bright images

(b) Dark images








Acknowledgment


References































Date post:	05-Oct-2021
Category:	Documents
Upload:	others
View:	2 times
Download:	0 times

Information Management and Applications of Intelligent ...

Documents