Research ArticleConcrete Spalling Detection for Metro Tunnel from Point CloudBased on Roughness Descriptor
HangbinWu Xingran Ao Zhuo Chen Chun Liu Zeran Xu and Pengfei Yu
College of Surveying and Geo-Informatics Tongji University Shanghai 200092 China
Correspondence should be addressed to Zhuo Chen czhuo0916tongjieducn
Received 20 January 2019 Revised 4 April 2019 Accepted 14 April 2019 Published 2 May 2019
Guest Editor Sang-Hoon Hong
Copyright copy 2019 Hangbin Wu 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
Automatic concrete spalling detection has become an important issue for metro tunnel examinations and maintenanceThis paperfocuses on concrete spalling detection research with surface roughness analysis based on point clouds produced by 3Dmobile laserscanning (MLS) system In the proposed method at first the points on ancillary facilities attached to tunnel surface are consideredas outliers and removed via circular scan-line fitting and large residual error filtering Then a roughness descriptor for the metrotunnel surface is designed based on the triangulated grid derived from point clouds The roughness descriptor is generally definedas the ratio of surface area to the projected area for a unit which works well in identifying high rough areas on the tunnel surfacesuch as bolt holes segment seams and spalling patches Finally rough area classification based on Hough transformation andsimilarity analysis is performed on the identified areas to accurately label patches belonging to segment seams and bolt holes Afterremoving the patches of bolt holes and segment seams the remaining patches are considered as belonging to concrete spallingTheexperiment was conducted on a real tunnel interval in Shanghai The result of concrete spalling detection revealed the validity andfeasibility of the proposed method
1 Introduction
Different from on-ground infrastructure tunnels are alwaysunder the complex environmental conditions and constantheavy traffic loads It is not uncommon to have damageon the tunnel surface due to the possibility of externalforces and material deterioration [1 2] Therefore duringthe service period of a tunnel regular inspection activitiesshould be carried out to check its health condition and regularmaintenance measures should be taken to keep its structuralintegrity and ensure the safety in the operation process [3]
A concrete spalling [4ndash6] as shown in Figure 1 is asmall but nonneglected tunnel damage which refers to thehappening of surface defects whose depths are deeper thannormal scaling caused by material deformation or fragiledeterioration Concrete spalling is one of the most seriousproblems that affects the performance and reliability of atunnel [7] Traditionally concrete spalling can be detectedby human visual inspection which needs tools of measuringtapes or profilers and to be identified by the size and location[8] However the process of manual inspection is time
consuming and with low efficiency and its result is alsosubjective and not reliable [9ndash12] Therefore it is urgent toreplace the traditional method with a more accurate andautomatic sensor-based method
Currently there are many researches focusing on detect-ing surface damage of concrete infrastructure with datagathered by different types of sensors These researches canbe generally classified in two ways The first and preferredway is the approach of visual imaging and analysing todetect concrete surface damage whose main advantagesare the contactless technique high speed in digital imageacquisition and the application of highly automated analysisprocedures For example Dawood et al [13] presented anintegrated framework for the detection and quantificationof concrete spalling distress from the digital images Theframework includes a hybrid algorithm for the detectionof concrete spalling regions interactive 3D presentationand regression analysis to estimate the relationship betweenspalling intensity and depth Medina et al [14] applied a newmethod called Gabor filter invariant to rotation allowing thedetection of cracks in any direction from the tunnel images
HindawiJournal of SensorsVolume 2019 Article ID 8574750 12 pageshttpsdoiorg10115520198574750
2 Journal of Sensors
Figure 1 Concrete spalling image
German et al [15] retrieved the spalling properties fromconcrete column images in an attempt to assess the safety ofpostearthquake concrete structures Koch and Brilakis [16]used histogram-based thresholds to segment the image intodefective and nondefective regions and then approximatedthe defective shapes using morphological refinement andelliptic regression Hutchinson et al [17] presented a sta-tistical method based on Bayesian decision theory for thepurpose of detecting concrete damage (cracks spalling etc)through conducting edge analysis of images However therequirement of illumination is essential for the camera toobtain high quality visual images which usually cannot besatisfied since there is no enough light in a tunnel and it ishard to find a long-time lasting power for the long rangetunnel inspection
The second class of laser scanning approach as an activedetection technique which can obtain high quality 3D pointcloud data even in the weak illumination has become moreand more popular in the civil engineering domain in recentyears [18ndash20] For example Teza et al [21] used datasetscollected by the terrestrial laser scanner to identify concretesurface damage using the mean curvature and Gaussiancurvature of the structure surface the application of whichmakes it possible to locate and quantify surface damageso as to enhance the current visual inspection strategyMizoguchi et al [22] evaluated the depth of scaling defectsbased on a customized region growing approach and iterativeclosest point (ICP) algorithm Liu et al [23] proposed asurface damage detection algorithm known as light detectionand ranging-based bridge evaluation (LiBE) for quantify-ing material quality loss The LiBE algorithm distinguishesinformation obtained from the original concrete surface bycalculating the surface gradient and displacement Tang etal [24] showed how laser scanners can be effectively usedto assess surface flatness and that it is possible to detectsurface flatness defects as small as 3 cm across and 1mmthick from a distance of 20m Yoon et al [25] proposeda method to detect cracks from laser scanned tunnel datausing radiometric and geometric properties of laser pointsNevertheless how to detect concrete spalling on the tunnelsurface has not been fully discussed This is because theeffective models for spalling detection mostly focused on theplanar concrete surface while the nonplanar tunnel surfaceis still not established which causes the concrete spalling intunnels to not be accurately detected
In this paper we propose a novel method that canautomatically detect the concrete spalling damage on tunnelsurface from 3D point cloud obtained by mobile laser scan-ning systemThe captured point cloud data not only containsinformation of the tunnel inner wall but also includes outlierpoints such as ancillary facilities and subway tracks Thusfirstly the outlier points need to be removed via circularscan-line fitting and large residual error filtering Then aroughness descriptor for themetro tunnel surface is designedbased on the triangulated grid derived from point cloudsThe roughness descriptor is generally defined as the ratio ofsurface area to the projected area for a unit which works wellin identifying high rough areas on the tunnel surface suchas bolt holes segment seams and spalling patches Finallyrough area classification based onHough transformation andsimilarity analysis is performed on the identified areas toaccurately label patches belonging to segment seams andbolt holes After removing the patches of bolt holes andseams the remaining patches are considered as belongingto concrete spalling Compared with previous studies theproposed method has the following characteristics (1) auto-matic concrete spalling detection for tunnel surfaces and (2)guidelines for choosing optimal scanning parameters
2 Research Methodology
21 Overview of the Proposed Method The proposed con-crete spalling detection method for metro tunnel based onroughness descriptor can be generally divided into threesteps namely outlier points removal roughness descriptorconstruction and rough area classification as shown inFigure 2 Point clouds used for concrete spalling detectionwere collected by a mobile three-dimensional laser scanningsystem The collected point cloud data not only contains thenecessary information of tunnel surface but also capturesthe unnecessary points such as cables pipelines and subwaytracks Therefore firstly these unnecessary outlier pointsincluding some derived from ancillary facilities and the otherfrom noise generated by the scanner will be removed onaccount of residual error filtering via a deviation thresholdThis part is introduced in Section 22
Secondly a roughness descriptor which is defined by theratio between surface area and projected area for a unit isconstructed and applied into the remaining points to evaluatethe rough situation of the tunnel surface To this end basedon the Poisson surface reconstruction method both theremaining points on tunnel surface and its correspondingprojected points on cylindrical surface are used to generatethe irregular triangulation and furthermore to calculatethe surface area and projected area respectively whosecalculation method is to find the sum of areas of first-order neighbourhood triangles around each point Thusthe roughness descriptor can be constructed and then thehigh rough areas including bolt holes segment seams andconcrete spalling on the tunnel surface can be identifiedThispart is given in Section 23
Finally a rough area classification algorithm is performedto separate bolt holes and segment seams from the rough
Journal of Sensors 3
Point clouds ofmetro tunnel
Triangle mesh constructionCylinder fitting
Surface area calculation
Triangle mesh construction
Projection area calculation
Concrete spalling
Roughness descriptor
Outliers removal
Classification algorithm
Figure 2 Flowchart of the proposed method
(a) (b) (c)
Figure 3 Three types of outlier points (a) Drift points (b) Redundant points (c) Mixed points
areas thus the remaining patches are considered as belongingto concrete spalling This is given in Section 24
22 Outlier Points Removal The point cloud used for con-crete spalling detection is collected by a mobile three-dimensional laser scanning system that scans the tunnelsurface in the form of a section during movementThereforethe captured point cloud data is stored as multiple scan lineswhich not only contains the information of tunnel surface
but also captures the outlier points mainly originating fromcables lighting equipment pipelines and other facilitiesattached to the inner wall subway tracks and noise generatedby the scanner These outliers however inevitably affectthe identification of concrete spalling and thus should beremoved at the very beginning In our method the outlierpoints are roughly grouped into three categories accordingto spatial distribution namely drift points redundant pointsand mixed points as shown in Figure 3
4 Journal of Sensors
The drift points are caused by the random noise gen-erated by the laser scanner whose spatial distribution ischaracterized as being dispersed and far away from themain part of metro tunnel Thus the drift points can beeliminated by the clustering algorithm In addition theredundant points come from subway track and other groundparts in metro tunnel and the mixed points stem fromsome ancillary facilities However the two types of points areusually mixed with nonoutlier points of the tunnel surfacethus it is necessary to adopt an effective method to realizefiltering
Since the tunnel cross-section is designed as a standardcircle the captured point clouds present in the form of alarge number of circular scanning lines According to this wepropose a filtering algorithm to remove redundant points andmixed points by identifying points with large residual errorEach circular scan line is first fitted with a circle model bythe RANSAC (random sample consistency) [26] in whichthe circlersquos boundary and centre are obtained After that theresidual error can be calculated as the distance between eachpoint on the scan line and the circlersquos boundary Finally thethreshold of residual error is set and the points with residualerrors larger than the threshold can be seen as outliers andremoved
23 Roughness Descriptor Construction From the perspec-tive of topography surface roughness refers to the unevennessof the ground generally defined as the ratio of the surfacearea to the projected area for a unit It is usually used toreflect the high and low undulations on the terrain thephenomenon of which has a similar shape to the undulationson the tunnel surface Therefore in this paper the concept ofsurface roughness is introduced into the metro tunnel for therough areas recognition
231 Surface Area and Projected Area Calculation A rough-ness descriptor needs to be constructed to identify the roughareas on tunnel surface According to the definition of surfaceroughness (the ratio of surface area to the projected area for aunit) in our method the surface area of a unit is representedby a polygon area enclosed by the points of the first-orderneighbourhood around each point Similarly points on thetunnel surface are projected onto the cylindrical surface andthe polygon area of the first-order neighbourhood aroundeach corresponding projected point is taken as the projectedarea of a unit
Since the tunnel surface is nonplanar in order to calculatethe surface area or the projected area around each point itis necessary to construct a triangular network for the pointcloud of tunnel as well as the corresponding projected pointcloud of cylinder respectively To this end Poisson surfacereconstruction [27] is adopted in this subsection whichis an intuitive method for mesh construction with pointcloud and its normal vector serving as input componentswhile the output manifests as a three-dimensional gridFigure 4 shows the modelling results of a part of tunnelpoint cloud and the details of triangular mesh Given therough situation on tunnel surface we do not calculate the
Figure 4 Triangular mesh of tunnel surface
Figure 5 Triangular mesh of cylindrical surface
area of the polygon enclosed by the points of the first-orderneighbourhood but the sum of triangular areas of the first-order neighbourhood around each point as the surface areaof a unit
What is more is that for the sake of calculating theprojected area the corresponding relationship of pointsbetween tunnel and cylinder needs to be established accu-rately Lei You et al [28] proposed an algorithm for projectingtrunk point clouds onto a cylindrical surface in sections toreconstruct the trunk surface the theory of which is alsoapplicable to the tunnel surface Based on the algorithm thetunnel surface can be defined by two parameters The firstone is centreline (L) that is described by a series of centrepoints (c119894(119888119909 119888119910 119888119911) 119894 = 1 sdot sdot sdot 119899)The second parameter is thedesign diameter (d) of a metro tunnel Taking into accountthe coordinate system of tunnel point cloud Z axis is locatedat the vertical scanning plane with upward direction positiveand both 119883 and 119884 axes are located at the lateral scanningplane and perpendicular to each other which forms a right-handed coordinate system wherein the positive directionof X axis points to the mileage direction Thus for anypoint 119901(119901119909119901119910119901119911) on the tunnel surface its correspondingprojection point 1199011015840 on the cylindrical surface satisfies
10038171003817100381710038171003817(1199011015840 minus (119888119910 119888119911 119901119909)) times (1 0 0)10038171003817100381710038171003817 minus 119889 = 0 (1)
where times is the outer product of vector and representsthe modulus of the vector It is noted that Equation (1)will have two solutions taking the point close to 119901 as theprojection point 1199011015840 on the cylindrical surface Figure 5 showsthe modelling results of a part of tunnel point cloud afterprojection and the details of triangularmeshThen accordingto the coordinates of projection points the polygon areaenclosed by the points of the first-order neighbourhoodaround each projection point is calculated as the projectedarea of a unit
Journal of Sensors 5
T
AB
D
E
F
G
normal area
rough area
Figure 6 Sketch of rough area points
232 Roughness Descriptor of Tunnel Surface In this paperwe define the ratio of surface area to the projected area aroundeach point as the roughness descriptor After constructing thetriangular mesh for the tunnel point cloud it is necessaryto calculate the area of each triangle and then find the sumof areas of first-order neighbourhood triangles around eachpoint which is considered as the surface area of a unitSimilarly point clouds on the cylindrical surface obtainedfrom projection also need to generate a triangulated gridwhere the polygon area enclosed by the points of the first-order neighbourhood for each point is performed as theprojected area of a unit
Simulating a set of points for a microelement on thetunnel surface as shown in Figure 6 the black points repre-sent normal area while the yellow zone represents the rougharea It is assumed that the red point T in the picture isin the rough area and other points marked with A(1199091 1199101)B(1199092 1199102) and D(1199093 1199103) etc are the normal points around TThe area of triangle which ismade up of point T and the othertwo points around T can be calculated by Heronrsquos formula[29] and stored Taking the triangle 119879119860119861 as an examplelengths of the corresponding three sides are represented as 119905119886 and 119887 respectively thus the area of which can be calculatedby
119878119879119860119861 = radic119902 (119902 minus 119905) (119902 minus 119886) (119902 minus 119887) q = (119905 + 119886 + 119887)2 (2)
Therefore the sum of triangle areas of the first-order neigh-bourhood around point T can be expressed as follows
Finally the roughness threshold should be set accurately toextract the points in high rough areas on tunnel surface
24 Rough Area Classification The points in rough areas onthe surface of metro tunnel can be extracted by the roughnessdescriptor successfully which are composed of three maincategories namely concrete spalling patches bolt holes andsegment seams However the three types of points extractedbased on the roughness descriptor are mixed together sowe need to separate the points belonging to the concretespalling patches from the rough points In our methodtaking into account the irregularity of spalling patches wecannot directly identify them from rough areas Accordingto this the method of rough area classification is adoptedto accurately label patches belonging to segment seams andbolt holes so that the remaining patches are considered asbelonging to concrete spalling
For the seams between segments if the tunnel point cloudis unfolded the seam appears as a straight line Thus in thispaper themethod for seam recognition is to project the pointcloud of metro tunnel onto a plane and rasterize it into animage After that the Hough transform [30] is applied torecognize the lines so that the seams of tunnel segments canbe determined and furthermore eliminated from the pointcloud
For the bolt holes there is a fixed size we can establish astandard point cloud template of bolt holes The point cloudis then clustered and the degree of similarity between eachsmall clustered group and the template is compared based onthe similarity analysismethod to determinewhich small clus-ter belongs to bolt hole In order to obtain the clustered pointcloud of bolt hole the mean-shift clustering algorithm [31] isapplied to the remaining point cloud after seam eliminationincluding bolt holes and spalling patches so that the pointclouds can form many different small groups Meanwhilewe established the point cloud library of bolt hole to beregarded as a template for recognition The size of bolt holewe used to collect the point cloud library is about 20lowast14lowast18(cm) and 17lowast15lowast18 (cm) Based on the similarity comparisonbetween the template and the clustered small groups boltholes can be identified from the point clouds The specificimplementationmethod is as follows Firstly it is necessary tostandardize the position of template points and each clusteredsmall group by performing PCA transformation (Principal
6 Journal of Sensors
Small group points
PCA transformationTemplate points
ICP registration
Feature vector calculation
Similarity calculation
Bolt hole
Principal componentsof small group points
Principal componentsof template points
Vector 1 Vector 2
Small group pointsafter registration
Figure 7 Procedure of bolt hole recognition
Component Analysis) [32] on the two three-dimensionalpoint clouds so that the three main components of bothare obtained and taken as the new standardized coordinatesystems Secondly after the coordinate transformation eachsmall group is registered with the template by using the ICP(Iterative Closest Point) [33] registration algorithm to furtheradjust the clustered points so as to have a similar postureto the template as much as possible Thirdly calculate thefeature vector of point cloud whose method is proposed byXiaotong H et al [34] for the principal components of thetemplate and the registered small group points respectivelyand furthermore perform similarity comparison between thetwo feature vectors to distinguish bolt holes from roughpoints Generally speaking any small group of point cloudswith similarity score greater than the accurate thresholdcan be identified as a bolt hole The procedure of bolt holerecognition is shown in Figure 7
25 Detectable Spalling Analysis In this paper we define theratio of surface area to the projected area around each pointas the roughness descriptor and simultaneously a formulathereof has also been deduced in Section 232 What is moreis that it is necessary to analyse theminimum spalling patches
that can be extracted using this method in metro tunnelTherefore it is assumed that Figure 6 shows a microelementon the tunnel surface under the ideal conditions wherethe point spacing is represented by 119898 The black point isin the normal area whose depth is zero while the redpoint is in the spalling area and the depth is ℎ Accordingto the roughness formula the sum of areas of first-orderneighbourhood triangles around point T can be expressed as
Hence one can see that in the position where the spallingdoes not occur or the nonrough position h=0 that is the
Journal of Sensors 7
Figure 8A scan line of tunnel point cloudTheblack ring representsthe real position of a section on the tunnel inner wall and thedeviation of red points away from black ring is regarded as theprecision of point cloud (Δ)
value of roughness descriptor is 1 When h gt 0 the value ofroughness descriptor is greater than 1 It indicates that theposition is rough relative to the normal position and may bespalling
Therefore the detectability of concrete spalling basedon the roughness descriptor is determined by the spallingdepth ℎ and the point spacingm while these two parametersare mainly affected by the instrument accuracy and theset parameters of the MLS system used when collectingpoint cloud in metro tunnel namely range error Δ and theresolution of laser scanner as well as the running velocityof MLS system Firstly the range error Δ of laser scannerindicates the precision of the collected point cloud of metrotunnel Taking out a scan line of point cloud and expanding itinto a straight line as shown by red points in Figure 8 assumethat the black ring is the real position of a section on thetunnel inner wall while the deviation of captured points awayfrom the innerwall is regarded as the precision of point cloudwhich is represented byΔThus it can be seen that the spallingpatches will not be detected when the value of spalling depthℎ is less than Δ
The other factor that affects the detectability of spallingis the point spacing m including the vertical spacing andlongitudinal spacing The vertical spacing of point clouddepends on the resolution of the scanner When settingdifferent resolutions the number of scanning points on theone scan-line changes accordingly In addition since thefrequency of scanner is usually fixed the velocity of themobile laser scanning system determines the point spacingin the direction of the mileage commonly referred to asthe longitudinal spacing Taking a microelement on thesurface of tunnel as an example in Figure 9 the verticaland longitudinal spacing of point cloud are represented by1198981 and 1198982 respectively and the blue areas are used toindicate the spalling patches It follows that when the areaof spalling patches less than the product of vertical spacingand longitudinal spacing it cannot be detected either
Therefore when the depth ℎ and the area 119878119888 of a concretespalling patch satisfy the following formula (9) it can beextracted from the point cloud of tunnel surface whichcan also be used as a guideline to select optimal scanningparameters for MLS system
ℎ gt 119878119888 gt 1198981 lowast 1198982 (9)
m1
m2
Figure 9 A microelement on tunnel surface The red dots indicatethe points on a microelement of tunnel surface captured by scannerand the blue areas indicate the spalling patches
Figure 10 Mobile laser scanning system
3 Case Study
31 Data Collection of Metro Tunnel A section of a metrotunnel in Shanghai was selected as the experimental area witha total length of about 250 meters The mobile laser scanning(MLS) system is equippedwith a scanner of FAROFOCUS3DX330 for point cloud data collection in the tunnel as shownin Figure 10 the scanner of which has a scanning range of300∘ and working frequency of 100Hz In order not to affectthe routine operation of the subway the experimental dataacquisition was carried out between midnight and three inthe morning And the resolution of scanner is set to 14 sothe number of points in one circular scan line is about 9760and the vertical point spacing 1198981 is about 2mm Generallyduring the period of data acquisition to ensure the densityof point cloud the running velocity of MLS system on thesubway track is set to 05ms so the average point spacing1198982 of the collected point clouds in the mileage direction isabout 5mm The general information of the case area anddata collection is shown in Table 1
32 Experimental Results
321 Outlier Removal Result of Tunnel Point Cloud Thecaptured point cloud data by MLS system mainly containsthe information of the tunnel surface where it is also mixedwith some outlier points originating from the subway trackscables lighting equipment and other facilities which willinevitably have great interference on the concrete spalling
8 Journal of Sensors
Resid
ual e
rror
Point serial number
004
002
0
minus002
minus004
minus006
minus008
minus01
minus0120 500 1000 1500 2000 2500 3000 3500
(a)
(b) (c)
Figure 11 Outlier removal (a) Residual error curve of one scan line (b) before the outlier removal and (c) after the outlier removal
Table 1 General information of the case area and data collection
Parameters ValueLength of case tunnel 250 mRadius of the tunnel 275 mAverage velocity during data collection 05 msScanning distance 330 mScanning range 300∘
Working frequency 100 HzResolution 14Range error () 2 mmVertical point spacing (1198981) 2 mmLongitudinal point spacing (1198982) 5 mmPoint density 100000 ptsm2
Total points gt 500000000 pts
identification Thus according to the outlier points removalalgorithm introduced in Section 22 the residual error curvefor each circular scan line can be generated an example ofwhich is shown in Figure 11(a) Since the fluctuation range ofresidual errors is between plus and minus 001 hence it canbe seen that points with an error of less than negative 001 canbe considered outliers and then eliminated Figures 11(b) and11(c) show the point cloud of tunnel before and after outlierpoints removal respectively
Most of the outlier points with a certain distance fromthe tunnel surface can be removed using the residual errorfiltering algorithm However there are still a small fractionof points from the bottom of pipeline facility that clings tothe tunnel inner wall and cannot be completely eliminatedthrough thismethod causing them to eventually be identifiedas rough areas This part of points is usually presented in theform of a line so it can be identified and further removedtogether with ring seams through the algorithm of Houghtransformation later
322 Rough Area Extraction After removing the outlierpoints from original point cloud data of metro tunnel aroughness descriptor based method is applied to extract thepoints of rough areas on tunnel surface for the purpose offurther identifying the concrete spalling patches therefromThus in order to calculate the value of roughness descriptorfor each point triangular meshes are first constructed forboth the remaining points after outlier removal and thecorresponding projected points on the cylindrical surfacethereby obtaining the surface area and projected area ofthe first-order neighbourhood around each point Thenaccording to the definition of roughness descriptor the ratioof surface area to the projected area around each point iscalculated a histogram of which is also generated indicatingthe number of points corresponding to different roughnessvalues as shown in Figure 12
Journal of Sensors 9
times105
poin
t num
ber
roughness descriptor
2
18
16
14
12
1
08
06
04
02
009 1 11 12 13 14 15 16 17
Figure 12 Histogram of roughness descriptor
Figure 13 Roughness map of tunnel surface
According to the value of the roughness descriptor foreach point obtained we need to determine an accuratethreshold to extract the points belonging to rough areasTaking into account the scanner accuracy and the set param-eters of MLS system during operation in this experimentmainly relying on the range error and the point spacing119898 the expression of roughness descriptor can be furtherrepresented as follows
It can be seen from formula (10) that when the value ofthe depth ℎ for a certain point tends to zero roughness ratiois at a critical condition Therefore with the value of pointspacing119898 and range error Δ in this case study the thresholdof roughness descriptor has been obtained as about 105 andpoints with proportion greater than 105 are filtered as roughareas Expand the rough point cloud on the tunnel surfaceinto a plane as shown in Figure 13
It should be noted that we cannot thin the original pointcloud otherwise some points belonging to rough areas maybe missed which will affect the detection of spalling damageHowever when calculating the roughness value of each pointthe number of tunnel point clouds captured byMLS system isvery huge which will take a lot of time to find a polygon area
surrounded by its first-order neighbourhood points for eachpoint Accordingly BitMap and BloomFilter are adopted inthis part to improve performing efficiency of the algorithmBitmap is a compact data storage structure that allocates 1 bitof memory for each element in the collection which greatlyreduces the storage space required to process massive pointcloud data Based on this structure BloomFilter completesthe query of the first-order neighbourhood points for eachpoint and then we can calculate the area of polygon enclosedby them which greatly compresses the memory space andshortens the calculation time
323 Results of Concrete Spalling Detection The rough areasextracted on tunnel surface based on the method of rough-ness descriptor mainly contain three types of objects namelyconcrete spalling patches bolt holes and the seams betweensegments In order to identify the points belonging to spallingpatches we used the method of rough area classificationdescribed in Section 24 to separate bolt holes and the seg-ment seams from rough points so that the concrete spallingpatches can be remained
Firstly if the tunnel surface is unfolded into a planethe seam appears as a line Therefore the method of seamidentification is to project the rough point cloud extractedonto a plane and rasterize it into an image Then the Houghtransformation algorithm can be used to identify seamsbetween segments the result of which is shown in Figure 14As we can see from the figure seams including the transverseseams longitudinal seams and oblique seams have beenidentified successfully
Then the bolt holes are detected using the similarityanalysis method the separation result of which is shown inFigure 14 where the bolt holes are shown in blue blocksThe remaining patches belong to the concrete spalling areashown in red After detecting the tunnel section of 250m itwas found through statistics that the spalling patches mainlyoccurred in the subinterval between 175 and 200m so that inorder to express the detected spalling more clearly Figure 14shows the information of tunnel between 175 and 200mFinally we found seven concrete spalling patches in this case
10 Journal of Sensors
Table 2 Basic information of concrete spalling patches
No Mileage [m] Spalling patches Images Spalling area [m2] Spalling depth [m]
J1 176 00191 0112
J2 176 00276 0126
J3 177 00377 0195
J4 178 00340 0099
J5 177 00384 0129
J6 177 00681 0083
J7 193 00249 0091
study the basic information of which is shown in Table 2and the mileage position corresponding to the spalling is alsogiven at the same time
According to formula (9) the theoretical depth value andarea value of minimum spalling patch are 2mm and 10mm2respectively and the extracted results are indeed greater thanthe theoretical minimum
In this paper in order to examine the accuracy of thedetected results we conducted a jointmanual inspectionwith
the maintenance company on the seven detected spallingpatches that is each of the spalling was confirmed one by onein the tunnel It was found that each spalling did occur at thecorresponding position Therefore the false detection rate iszero and it is confirmed that the proposed concrete spallingdetection algorithm performs well In addition taking intoaccount the small size of the spalling and the large spacingbetween the points it does cause omission errors which havebeen analysed in Section 25 However these regions that
Journal of Sensors 11
Bolt holeConcrete spalling
Segment seam
7
6 5 4
32
1
Figure 14 Recognition result of bolt holes segment seams andconcrete spalling patches in tunnel subinterval of 175-200 m
could not be detected are very small and belong to the normalcategory which will not affect the performance and reliabilityof a tunnel
4 Conclusion
In this paper a new method used for concrete spallingdetection in metro tunnel from point cloud based on theroughness descriptor is proposed Firstly the point cloudacquired by mobile laser scanning system needs to eliminateoutlier points originating from ancillary facilities attachedto shield tunnel wall based on the residual error filteringalgorithmThen a roughness descriptor for the metro tunnelsurface is designed to extract the rough areas on the tunnelsurface including bolt holes segment seams and spallingpatches Finally rough area classification is performed onthe identified rough areas to accurately separate the segmentseams and bolt holes from rough areas so that the concretespalling patches are left A section of metro tunnel intervalabout 250m in Shanghai is selected to verify the validityof the proposed method and seven concrete spalling areasare detected which are identified as surface defects in metrotunnel This could be helpful for tunnel maintenance andoperation safety Compared with previous studies the con-cept of roughness descriptor is proposed to detect concretespalling which is suitable for not only flat concrete surfacesbut also nonplanar concrete surfaces and at the same timeoffer the guidance for optimal scanning parameter selection
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This research is supported by the National Science Foun-dation of China (no 41671451) the National Science andTechnologyMajor Program (2016YFB0502104) and the Fun-damental Research Funds for the Central Universities ofChina The authors would like to express appreciation tocolleagues in our laboratory for their valuable commentshelp
References
[1] T Asakura and Y Kojima ldquoTunnel maintenance in JapanrdquoTunnelling and Underground Space Technology vol 18 no 2-3pp 161ndash169 2003
[2] Y Yuan Y Bai and J Liu ldquoAssessment service state of tunnelstructurerdquo Tunnelling and Underground Space Technology vol27 no 1 pp 72ndash85 2012
[3] F Sandrone and V Labiouse ldquoIdentification and analysisof Swiss National Road tunnels pathologiesrdquo Tunnelling andUnderground Space Technology vol 26 no 2 pp 374ndash390 2011
[4] Portland Cement Association (PCA) Concrete slab surfacedefects Causes Prevention And Repair Portland CementSkokie IL USA 2001
[5] N Delatte S Chen N Maini et al ldquoApplication of non-destructive evaluation to subway tunnel systemsrdquo Transporta-tion Research Record vol 1845 no 3 pp 127ndash135 2003
[6] H Russell and J Gilmore ldquoInspection policy and proceduresfor rail transit tunnels and underground structuresrdquo TransitCooperative Research Program Synthesis of Transit Practice 1997
[7] Y Yuan X Jiang and Q Ai ldquoProbabilistic assessment forconcrete spalling in tunnel structuresrdquo ASCE-ASME Journalof Risk and Uncertainty in Engineering Systems Part A CivilEngineering vol 3 no 4 2017
[8] M-K Kim H Sohn and C-C Chang ldquoLocalization andquantification of concrete spalling defects using terrestrial laserscanningrdquo Journal of Computing in Civil Engineering vol 29 no6 2015
[9] Z Zhu and I Bilakis ldquoDetecting air pockets for architecturalconcrete quality assessment using visual sensingrdquo ElectronicJournal of Information Technology in Construction vol 13 pp86ndash102 2008
[10] American Concrete Institute ACI Manual of Concrete Inspec-tion ACI Committee 311 SP-2(07) Detroit Mich USA 2007
[11] A M Paterson G R Dowling and D A ChamberlainldquoBuilding inspection can computer vision helprdquo Automationin Construction vol 7 no 1 pp 13ndash20 1997
[12] B Guldur Erkal and J F Hajjar ldquoLaser-based surface damagedetection and quantification using predicted surface proper-tiesrdquo Automation in Construction vol 83 pp 285ndash302 2017
[13] T Dawood Z Zhu and T Zayed ldquoMachine vision-basedmodelfor spalling detection and quantification in subway networksrdquoAutomation in Construction vol 81 pp 149ndash160 2017
[14] RMedina J Llamas J Gomez-Garcıa-Bermejo E Zalama andM Segarra ldquoCrack detection in concrete tunnels using a Gaborfilter invariant to rotationrdquo Sensors vol 17 no 7 p 1670 2017
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
2 Journal of Sensors
Figure 1 Concrete spalling image
German et al [15] retrieved the spalling properties fromconcrete column images in an attempt to assess the safety ofpostearthquake concrete structures Koch and Brilakis [16]used histogram-based thresholds to segment the image intodefective and nondefective regions and then approximatedthe defective shapes using morphological refinement andelliptic regression Hutchinson et al [17] presented a sta-tistical method based on Bayesian decision theory for thepurpose of detecting concrete damage (cracks spalling etc)through conducting edge analysis of images However therequirement of illumination is essential for the camera toobtain high quality visual images which usually cannot besatisfied since there is no enough light in a tunnel and it ishard to find a long-time lasting power for the long rangetunnel inspection
The second class of laser scanning approach as an activedetection technique which can obtain high quality 3D pointcloud data even in the weak illumination has become moreand more popular in the civil engineering domain in recentyears [18ndash20] For example Teza et al [21] used datasetscollected by the terrestrial laser scanner to identify concretesurface damage using the mean curvature and Gaussiancurvature of the structure surface the application of whichmakes it possible to locate and quantify surface damageso as to enhance the current visual inspection strategyMizoguchi et al [22] evaluated the depth of scaling defectsbased on a customized region growing approach and iterativeclosest point (ICP) algorithm Liu et al [23] proposed asurface damage detection algorithm known as light detectionand ranging-based bridge evaluation (LiBE) for quantify-ing material quality loss The LiBE algorithm distinguishesinformation obtained from the original concrete surface bycalculating the surface gradient and displacement Tang etal [24] showed how laser scanners can be effectively usedto assess surface flatness and that it is possible to detectsurface flatness defects as small as 3 cm across and 1mmthick from a distance of 20m Yoon et al [25] proposeda method to detect cracks from laser scanned tunnel datausing radiometric and geometric properties of laser pointsNevertheless how to detect concrete spalling on the tunnelsurface has not been fully discussed This is because theeffective models for spalling detection mostly focused on theplanar concrete surface while the nonplanar tunnel surfaceis still not established which causes the concrete spalling intunnels to not be accurately detected
In this paper we propose a novel method that canautomatically detect the concrete spalling damage on tunnelsurface from 3D point cloud obtained by mobile laser scan-ning systemThe captured point cloud data not only containsinformation of the tunnel inner wall but also includes outlierpoints such as ancillary facilities and subway tracks Thusfirstly the outlier points need to be removed via circularscan-line fitting and large residual error filtering Then aroughness descriptor for themetro tunnel surface is designedbased on the triangulated grid derived from point cloudsThe roughness descriptor is generally defined as the ratio ofsurface area to the projected area for a unit which works wellin identifying high rough areas on the tunnel surface suchas bolt holes segment seams and spalling patches Finallyrough area classification based onHough transformation andsimilarity analysis is performed on the identified areas toaccurately label patches belonging to segment seams andbolt holes After removing the patches of bolt holes andseams the remaining patches are considered as belongingto concrete spalling Compared with previous studies theproposed method has the following characteristics (1) auto-matic concrete spalling detection for tunnel surfaces and (2)guidelines for choosing optimal scanning parameters
2 Research Methodology
21 Overview of the Proposed Method The proposed con-crete spalling detection method for metro tunnel based onroughness descriptor can be generally divided into threesteps namely outlier points removal roughness descriptorconstruction and rough area classification as shown inFigure 2 Point clouds used for concrete spalling detectionwere collected by a mobile three-dimensional laser scanningsystem The collected point cloud data not only contains thenecessary information of tunnel surface but also capturesthe unnecessary points such as cables pipelines and subwaytracks Therefore firstly these unnecessary outlier pointsincluding some derived from ancillary facilities and the otherfrom noise generated by the scanner will be removed onaccount of residual error filtering via a deviation thresholdThis part is introduced in Section 22
Secondly a roughness descriptor which is defined by theratio between surface area and projected area for a unit isconstructed and applied into the remaining points to evaluatethe rough situation of the tunnel surface To this end basedon the Poisson surface reconstruction method both theremaining points on tunnel surface and its correspondingprojected points on cylindrical surface are used to generatethe irregular triangulation and furthermore to calculatethe surface area and projected area respectively whosecalculation method is to find the sum of areas of first-order neighbourhood triangles around each point Thusthe roughness descriptor can be constructed and then thehigh rough areas including bolt holes segment seams andconcrete spalling on the tunnel surface can be identifiedThispart is given in Section 23
Finally a rough area classification algorithm is performedto separate bolt holes and segment seams from the rough
Journal of Sensors 3
Point clouds ofmetro tunnel
Triangle mesh constructionCylinder fitting
Surface area calculation
Triangle mesh construction
Projection area calculation
Concrete spalling
Roughness descriptor
Outliers removal
Classification algorithm
Figure 2 Flowchart of the proposed method
(a) (b) (c)
Figure 3 Three types of outlier points (a) Drift points (b) Redundant points (c) Mixed points
areas thus the remaining patches are considered as belongingto concrete spalling This is given in Section 24
22 Outlier Points Removal The point cloud used for con-crete spalling detection is collected by a mobile three-dimensional laser scanning system that scans the tunnelsurface in the form of a section during movementThereforethe captured point cloud data is stored as multiple scan lineswhich not only contains the information of tunnel surface
but also captures the outlier points mainly originating fromcables lighting equipment pipelines and other facilitiesattached to the inner wall subway tracks and noise generatedby the scanner These outliers however inevitably affectthe identification of concrete spalling and thus should beremoved at the very beginning In our method the outlierpoints are roughly grouped into three categories accordingto spatial distribution namely drift points redundant pointsand mixed points as shown in Figure 3
4 Journal of Sensors
The drift points are caused by the random noise gen-erated by the laser scanner whose spatial distribution ischaracterized as being dispersed and far away from themain part of metro tunnel Thus the drift points can beeliminated by the clustering algorithm In addition theredundant points come from subway track and other groundparts in metro tunnel and the mixed points stem fromsome ancillary facilities However the two types of points areusually mixed with nonoutlier points of the tunnel surfacethus it is necessary to adopt an effective method to realizefiltering
Since the tunnel cross-section is designed as a standardcircle the captured point clouds present in the form of alarge number of circular scanning lines According to this wepropose a filtering algorithm to remove redundant points andmixed points by identifying points with large residual errorEach circular scan line is first fitted with a circle model bythe RANSAC (random sample consistency) [26] in whichthe circlersquos boundary and centre are obtained After that theresidual error can be calculated as the distance between eachpoint on the scan line and the circlersquos boundary Finally thethreshold of residual error is set and the points with residualerrors larger than the threshold can be seen as outliers andremoved
23 Roughness Descriptor Construction From the perspec-tive of topography surface roughness refers to the unevennessof the ground generally defined as the ratio of the surfacearea to the projected area for a unit It is usually used toreflect the high and low undulations on the terrain thephenomenon of which has a similar shape to the undulationson the tunnel surface Therefore in this paper the concept ofsurface roughness is introduced into the metro tunnel for therough areas recognition
231 Surface Area and Projected Area Calculation A rough-ness descriptor needs to be constructed to identify the roughareas on tunnel surface According to the definition of surfaceroughness (the ratio of surface area to the projected area for aunit) in our method the surface area of a unit is representedby a polygon area enclosed by the points of the first-orderneighbourhood around each point Similarly points on thetunnel surface are projected onto the cylindrical surface andthe polygon area of the first-order neighbourhood aroundeach corresponding projected point is taken as the projectedarea of a unit
Since the tunnel surface is nonplanar in order to calculatethe surface area or the projected area around each point itis necessary to construct a triangular network for the pointcloud of tunnel as well as the corresponding projected pointcloud of cylinder respectively To this end Poisson surfacereconstruction [27] is adopted in this subsection whichis an intuitive method for mesh construction with pointcloud and its normal vector serving as input componentswhile the output manifests as a three-dimensional gridFigure 4 shows the modelling results of a part of tunnelpoint cloud and the details of triangular mesh Given therough situation on tunnel surface we do not calculate the
Figure 4 Triangular mesh of tunnel surface
Figure 5 Triangular mesh of cylindrical surface
area of the polygon enclosed by the points of the first-orderneighbourhood but the sum of triangular areas of the first-order neighbourhood around each point as the surface areaof a unit
What is more is that for the sake of calculating theprojected area the corresponding relationship of pointsbetween tunnel and cylinder needs to be established accu-rately Lei You et al [28] proposed an algorithm for projectingtrunk point clouds onto a cylindrical surface in sections toreconstruct the trunk surface the theory of which is alsoapplicable to the tunnel surface Based on the algorithm thetunnel surface can be defined by two parameters The firstone is centreline (L) that is described by a series of centrepoints (c119894(119888119909 119888119910 119888119911) 119894 = 1 sdot sdot sdot 119899)The second parameter is thedesign diameter (d) of a metro tunnel Taking into accountthe coordinate system of tunnel point cloud Z axis is locatedat the vertical scanning plane with upward direction positiveand both 119883 and 119884 axes are located at the lateral scanningplane and perpendicular to each other which forms a right-handed coordinate system wherein the positive directionof X axis points to the mileage direction Thus for anypoint 119901(119901119909119901119910119901119911) on the tunnel surface its correspondingprojection point 1199011015840 on the cylindrical surface satisfies
10038171003817100381710038171003817(1199011015840 minus (119888119910 119888119911 119901119909)) times (1 0 0)10038171003817100381710038171003817 minus 119889 = 0 (1)
where times is the outer product of vector and representsthe modulus of the vector It is noted that Equation (1)will have two solutions taking the point close to 119901 as theprojection point 1199011015840 on the cylindrical surface Figure 5 showsthe modelling results of a part of tunnel point cloud afterprojection and the details of triangularmeshThen accordingto the coordinates of projection points the polygon areaenclosed by the points of the first-order neighbourhoodaround each projection point is calculated as the projectedarea of a unit
Journal of Sensors 5
T
AB
D
E
F
G
normal area
rough area
Figure 6 Sketch of rough area points
232 Roughness Descriptor of Tunnel Surface In this paperwe define the ratio of surface area to the projected area aroundeach point as the roughness descriptor After constructing thetriangular mesh for the tunnel point cloud it is necessaryto calculate the area of each triangle and then find the sumof areas of first-order neighbourhood triangles around eachpoint which is considered as the surface area of a unitSimilarly point clouds on the cylindrical surface obtainedfrom projection also need to generate a triangulated gridwhere the polygon area enclosed by the points of the first-order neighbourhood for each point is performed as theprojected area of a unit
Simulating a set of points for a microelement on thetunnel surface as shown in Figure 6 the black points repre-sent normal area while the yellow zone represents the rougharea It is assumed that the red point T in the picture isin the rough area and other points marked with A(1199091 1199101)B(1199092 1199102) and D(1199093 1199103) etc are the normal points around TThe area of triangle which ismade up of point T and the othertwo points around T can be calculated by Heronrsquos formula[29] and stored Taking the triangle 119879119860119861 as an examplelengths of the corresponding three sides are represented as 119905119886 and 119887 respectively thus the area of which can be calculatedby
119878119879119860119861 = radic119902 (119902 minus 119905) (119902 minus 119886) (119902 minus 119887) q = (119905 + 119886 + 119887)2 (2)
Therefore the sum of triangle areas of the first-order neigh-bourhood around point T can be expressed as follows
Finally the roughness threshold should be set accurately toextract the points in high rough areas on tunnel surface
24 Rough Area Classification The points in rough areas onthe surface of metro tunnel can be extracted by the roughnessdescriptor successfully which are composed of three maincategories namely concrete spalling patches bolt holes andsegment seams However the three types of points extractedbased on the roughness descriptor are mixed together sowe need to separate the points belonging to the concretespalling patches from the rough points In our methodtaking into account the irregularity of spalling patches wecannot directly identify them from rough areas Accordingto this the method of rough area classification is adoptedto accurately label patches belonging to segment seams andbolt holes so that the remaining patches are considered asbelonging to concrete spalling
For the seams between segments if the tunnel point cloudis unfolded the seam appears as a straight line Thus in thispaper themethod for seam recognition is to project the pointcloud of metro tunnel onto a plane and rasterize it into animage After that the Hough transform [30] is applied torecognize the lines so that the seams of tunnel segments canbe determined and furthermore eliminated from the pointcloud
For the bolt holes there is a fixed size we can establish astandard point cloud template of bolt holes The point cloudis then clustered and the degree of similarity between eachsmall clustered group and the template is compared based onthe similarity analysismethod to determinewhich small clus-ter belongs to bolt hole In order to obtain the clustered pointcloud of bolt hole the mean-shift clustering algorithm [31] isapplied to the remaining point cloud after seam eliminationincluding bolt holes and spalling patches so that the pointclouds can form many different small groups Meanwhilewe established the point cloud library of bolt hole to beregarded as a template for recognition The size of bolt holewe used to collect the point cloud library is about 20lowast14lowast18(cm) and 17lowast15lowast18 (cm) Based on the similarity comparisonbetween the template and the clustered small groups boltholes can be identified from the point clouds The specificimplementationmethod is as follows Firstly it is necessary tostandardize the position of template points and each clusteredsmall group by performing PCA transformation (Principal
6 Journal of Sensors
Small group points
PCA transformationTemplate points
ICP registration
Feature vector calculation
Similarity calculation
Bolt hole
Principal componentsof small group points
Principal componentsof template points
Vector 1 Vector 2
Small group pointsafter registration
Figure 7 Procedure of bolt hole recognition
Component Analysis) [32] on the two three-dimensionalpoint clouds so that the three main components of bothare obtained and taken as the new standardized coordinatesystems Secondly after the coordinate transformation eachsmall group is registered with the template by using the ICP(Iterative Closest Point) [33] registration algorithm to furtheradjust the clustered points so as to have a similar postureto the template as much as possible Thirdly calculate thefeature vector of point cloud whose method is proposed byXiaotong H et al [34] for the principal components of thetemplate and the registered small group points respectivelyand furthermore perform similarity comparison between thetwo feature vectors to distinguish bolt holes from roughpoints Generally speaking any small group of point cloudswith similarity score greater than the accurate thresholdcan be identified as a bolt hole The procedure of bolt holerecognition is shown in Figure 7
25 Detectable Spalling Analysis In this paper we define theratio of surface area to the projected area around each pointas the roughness descriptor and simultaneously a formulathereof has also been deduced in Section 232 What is moreis that it is necessary to analyse theminimum spalling patches
that can be extracted using this method in metro tunnelTherefore it is assumed that Figure 6 shows a microelementon the tunnel surface under the ideal conditions wherethe point spacing is represented by 119898 The black point isin the normal area whose depth is zero while the redpoint is in the spalling area and the depth is ℎ Accordingto the roughness formula the sum of areas of first-orderneighbourhood triangles around point T can be expressed as
Hence one can see that in the position where the spallingdoes not occur or the nonrough position h=0 that is the
Journal of Sensors 7
Figure 8A scan line of tunnel point cloudTheblack ring representsthe real position of a section on the tunnel inner wall and thedeviation of red points away from black ring is regarded as theprecision of point cloud (Δ)
value of roughness descriptor is 1 When h gt 0 the value ofroughness descriptor is greater than 1 It indicates that theposition is rough relative to the normal position and may bespalling
Therefore the detectability of concrete spalling basedon the roughness descriptor is determined by the spallingdepth ℎ and the point spacingm while these two parametersare mainly affected by the instrument accuracy and theset parameters of the MLS system used when collectingpoint cloud in metro tunnel namely range error Δ and theresolution of laser scanner as well as the running velocityof MLS system Firstly the range error Δ of laser scannerindicates the precision of the collected point cloud of metrotunnel Taking out a scan line of point cloud and expanding itinto a straight line as shown by red points in Figure 8 assumethat the black ring is the real position of a section on thetunnel inner wall while the deviation of captured points awayfrom the innerwall is regarded as the precision of point cloudwhich is represented byΔThus it can be seen that the spallingpatches will not be detected when the value of spalling depthℎ is less than Δ
The other factor that affects the detectability of spallingis the point spacing m including the vertical spacing andlongitudinal spacing The vertical spacing of point clouddepends on the resolution of the scanner When settingdifferent resolutions the number of scanning points on theone scan-line changes accordingly In addition since thefrequency of scanner is usually fixed the velocity of themobile laser scanning system determines the point spacingin the direction of the mileage commonly referred to asthe longitudinal spacing Taking a microelement on thesurface of tunnel as an example in Figure 9 the verticaland longitudinal spacing of point cloud are represented by1198981 and 1198982 respectively and the blue areas are used toindicate the spalling patches It follows that when the areaof spalling patches less than the product of vertical spacingand longitudinal spacing it cannot be detected either
Therefore when the depth ℎ and the area 119878119888 of a concretespalling patch satisfy the following formula (9) it can beextracted from the point cloud of tunnel surface whichcan also be used as a guideline to select optimal scanningparameters for MLS system
ℎ gt 119878119888 gt 1198981 lowast 1198982 (9)
m1
m2
Figure 9 A microelement on tunnel surface The red dots indicatethe points on a microelement of tunnel surface captured by scannerand the blue areas indicate the spalling patches
Figure 10 Mobile laser scanning system
3 Case Study
31 Data Collection of Metro Tunnel A section of a metrotunnel in Shanghai was selected as the experimental area witha total length of about 250 meters The mobile laser scanning(MLS) system is equippedwith a scanner of FAROFOCUS3DX330 for point cloud data collection in the tunnel as shownin Figure 10 the scanner of which has a scanning range of300∘ and working frequency of 100Hz In order not to affectthe routine operation of the subway the experimental dataacquisition was carried out between midnight and three inthe morning And the resolution of scanner is set to 14 sothe number of points in one circular scan line is about 9760and the vertical point spacing 1198981 is about 2mm Generallyduring the period of data acquisition to ensure the densityof point cloud the running velocity of MLS system on thesubway track is set to 05ms so the average point spacing1198982 of the collected point clouds in the mileage direction isabout 5mm The general information of the case area anddata collection is shown in Table 1
32 Experimental Results
321 Outlier Removal Result of Tunnel Point Cloud Thecaptured point cloud data by MLS system mainly containsthe information of the tunnel surface where it is also mixedwith some outlier points originating from the subway trackscables lighting equipment and other facilities which willinevitably have great interference on the concrete spalling
8 Journal of Sensors
Resid
ual e
rror
Point serial number
004
002
0
minus002
minus004
minus006
minus008
minus01
minus0120 500 1000 1500 2000 2500 3000 3500
(a)
(b) (c)
Figure 11 Outlier removal (a) Residual error curve of one scan line (b) before the outlier removal and (c) after the outlier removal
Table 1 General information of the case area and data collection
Parameters ValueLength of case tunnel 250 mRadius of the tunnel 275 mAverage velocity during data collection 05 msScanning distance 330 mScanning range 300∘
Working frequency 100 HzResolution 14Range error () 2 mmVertical point spacing (1198981) 2 mmLongitudinal point spacing (1198982) 5 mmPoint density 100000 ptsm2
Total points gt 500000000 pts
identification Thus according to the outlier points removalalgorithm introduced in Section 22 the residual error curvefor each circular scan line can be generated an example ofwhich is shown in Figure 11(a) Since the fluctuation range ofresidual errors is between plus and minus 001 hence it canbe seen that points with an error of less than negative 001 canbe considered outliers and then eliminated Figures 11(b) and11(c) show the point cloud of tunnel before and after outlierpoints removal respectively
Most of the outlier points with a certain distance fromthe tunnel surface can be removed using the residual errorfiltering algorithm However there are still a small fractionof points from the bottom of pipeline facility that clings tothe tunnel inner wall and cannot be completely eliminatedthrough thismethod causing them to eventually be identifiedas rough areas This part of points is usually presented in theform of a line so it can be identified and further removedtogether with ring seams through the algorithm of Houghtransformation later
322 Rough Area Extraction After removing the outlierpoints from original point cloud data of metro tunnel aroughness descriptor based method is applied to extract thepoints of rough areas on tunnel surface for the purpose offurther identifying the concrete spalling patches therefromThus in order to calculate the value of roughness descriptorfor each point triangular meshes are first constructed forboth the remaining points after outlier removal and thecorresponding projected points on the cylindrical surfacethereby obtaining the surface area and projected area ofthe first-order neighbourhood around each point Thenaccording to the definition of roughness descriptor the ratioof surface area to the projected area around each point iscalculated a histogram of which is also generated indicatingthe number of points corresponding to different roughnessvalues as shown in Figure 12
Journal of Sensors 9
times105
poin
t num
ber
roughness descriptor
2
18
16
14
12
1
08
06
04
02
009 1 11 12 13 14 15 16 17
Figure 12 Histogram of roughness descriptor
Figure 13 Roughness map of tunnel surface
According to the value of the roughness descriptor foreach point obtained we need to determine an accuratethreshold to extract the points belonging to rough areasTaking into account the scanner accuracy and the set param-eters of MLS system during operation in this experimentmainly relying on the range error and the point spacing119898 the expression of roughness descriptor can be furtherrepresented as follows
It can be seen from formula (10) that when the value ofthe depth ℎ for a certain point tends to zero roughness ratiois at a critical condition Therefore with the value of pointspacing119898 and range error Δ in this case study the thresholdof roughness descriptor has been obtained as about 105 andpoints with proportion greater than 105 are filtered as roughareas Expand the rough point cloud on the tunnel surfaceinto a plane as shown in Figure 13
It should be noted that we cannot thin the original pointcloud otherwise some points belonging to rough areas maybe missed which will affect the detection of spalling damageHowever when calculating the roughness value of each pointthe number of tunnel point clouds captured byMLS system isvery huge which will take a lot of time to find a polygon area
surrounded by its first-order neighbourhood points for eachpoint Accordingly BitMap and BloomFilter are adopted inthis part to improve performing efficiency of the algorithmBitmap is a compact data storage structure that allocates 1 bitof memory for each element in the collection which greatlyreduces the storage space required to process massive pointcloud data Based on this structure BloomFilter completesthe query of the first-order neighbourhood points for eachpoint and then we can calculate the area of polygon enclosedby them which greatly compresses the memory space andshortens the calculation time
323 Results of Concrete Spalling Detection The rough areasextracted on tunnel surface based on the method of rough-ness descriptor mainly contain three types of objects namelyconcrete spalling patches bolt holes and the seams betweensegments In order to identify the points belonging to spallingpatches we used the method of rough area classificationdescribed in Section 24 to separate bolt holes and the seg-ment seams from rough points so that the concrete spallingpatches can be remained
Firstly if the tunnel surface is unfolded into a planethe seam appears as a line Therefore the method of seamidentification is to project the rough point cloud extractedonto a plane and rasterize it into an image Then the Houghtransformation algorithm can be used to identify seamsbetween segments the result of which is shown in Figure 14As we can see from the figure seams including the transverseseams longitudinal seams and oblique seams have beenidentified successfully
Then the bolt holes are detected using the similarityanalysis method the separation result of which is shown inFigure 14 where the bolt holes are shown in blue blocksThe remaining patches belong to the concrete spalling areashown in red After detecting the tunnel section of 250m itwas found through statistics that the spalling patches mainlyoccurred in the subinterval between 175 and 200m so that inorder to express the detected spalling more clearly Figure 14shows the information of tunnel between 175 and 200mFinally we found seven concrete spalling patches in this case
10 Journal of Sensors
Table 2 Basic information of concrete spalling patches
No Mileage [m] Spalling patches Images Spalling area [m2] Spalling depth [m]
J1 176 00191 0112
J2 176 00276 0126
J3 177 00377 0195
J4 178 00340 0099
J5 177 00384 0129
J6 177 00681 0083
J7 193 00249 0091
study the basic information of which is shown in Table 2and the mileage position corresponding to the spalling is alsogiven at the same time
According to formula (9) the theoretical depth value andarea value of minimum spalling patch are 2mm and 10mm2respectively and the extracted results are indeed greater thanthe theoretical minimum
In this paper in order to examine the accuracy of thedetected results we conducted a jointmanual inspectionwith
the maintenance company on the seven detected spallingpatches that is each of the spalling was confirmed one by onein the tunnel It was found that each spalling did occur at thecorresponding position Therefore the false detection rate iszero and it is confirmed that the proposed concrete spallingdetection algorithm performs well In addition taking intoaccount the small size of the spalling and the large spacingbetween the points it does cause omission errors which havebeen analysed in Section 25 However these regions that
Journal of Sensors 11
Bolt holeConcrete spalling
Segment seam
7
6 5 4
32
1
Figure 14 Recognition result of bolt holes segment seams andconcrete spalling patches in tunnel subinterval of 175-200 m
could not be detected are very small and belong to the normalcategory which will not affect the performance and reliabilityof a tunnel
4 Conclusion
In this paper a new method used for concrete spallingdetection in metro tunnel from point cloud based on theroughness descriptor is proposed Firstly the point cloudacquired by mobile laser scanning system needs to eliminateoutlier points originating from ancillary facilities attachedto shield tunnel wall based on the residual error filteringalgorithmThen a roughness descriptor for the metro tunnelsurface is designed to extract the rough areas on the tunnelsurface including bolt holes segment seams and spallingpatches Finally rough area classification is performed onthe identified rough areas to accurately separate the segmentseams and bolt holes from rough areas so that the concretespalling patches are left A section of metro tunnel intervalabout 250m in Shanghai is selected to verify the validityof the proposed method and seven concrete spalling areasare detected which are identified as surface defects in metrotunnel This could be helpful for tunnel maintenance andoperation safety Compared with previous studies the con-cept of roughness descriptor is proposed to detect concretespalling which is suitable for not only flat concrete surfacesbut also nonplanar concrete surfaces and at the same timeoffer the guidance for optimal scanning parameter selection
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This research is supported by the National Science Foun-dation of China (no 41671451) the National Science andTechnologyMajor Program (2016YFB0502104) and the Fun-damental Research Funds for the Central Universities ofChina The authors would like to express appreciation tocolleagues in our laboratory for their valuable commentshelp
References
[1] T Asakura and Y Kojima ldquoTunnel maintenance in JapanrdquoTunnelling and Underground Space Technology vol 18 no 2-3pp 161ndash169 2003
[2] Y Yuan Y Bai and J Liu ldquoAssessment service state of tunnelstructurerdquo Tunnelling and Underground Space Technology vol27 no 1 pp 72ndash85 2012
[3] F Sandrone and V Labiouse ldquoIdentification and analysisof Swiss National Road tunnels pathologiesrdquo Tunnelling andUnderground Space Technology vol 26 no 2 pp 374ndash390 2011
[4] Portland Cement Association (PCA) Concrete slab surfacedefects Causes Prevention And Repair Portland CementSkokie IL USA 2001
[5] N Delatte S Chen N Maini et al ldquoApplication of non-destructive evaluation to subway tunnel systemsrdquo Transporta-tion Research Record vol 1845 no 3 pp 127ndash135 2003
[6] H Russell and J Gilmore ldquoInspection policy and proceduresfor rail transit tunnels and underground structuresrdquo TransitCooperative Research Program Synthesis of Transit Practice 1997
[7] Y Yuan X Jiang and Q Ai ldquoProbabilistic assessment forconcrete spalling in tunnel structuresrdquo ASCE-ASME Journalof Risk and Uncertainty in Engineering Systems Part A CivilEngineering vol 3 no 4 2017
[8] M-K Kim H Sohn and C-C Chang ldquoLocalization andquantification of concrete spalling defects using terrestrial laserscanningrdquo Journal of Computing in Civil Engineering vol 29 no6 2015
[9] Z Zhu and I Bilakis ldquoDetecting air pockets for architecturalconcrete quality assessment using visual sensingrdquo ElectronicJournal of Information Technology in Construction vol 13 pp86ndash102 2008
[10] American Concrete Institute ACI Manual of Concrete Inspec-tion ACI Committee 311 SP-2(07) Detroit Mich USA 2007
[11] A M Paterson G R Dowling and D A ChamberlainldquoBuilding inspection can computer vision helprdquo Automationin Construction vol 7 no 1 pp 13ndash20 1997
[12] B Guldur Erkal and J F Hajjar ldquoLaser-based surface damagedetection and quantification using predicted surface proper-tiesrdquo Automation in Construction vol 83 pp 285ndash302 2017
[13] T Dawood Z Zhu and T Zayed ldquoMachine vision-basedmodelfor spalling detection and quantification in subway networksrdquoAutomation in Construction vol 81 pp 149ndash160 2017
[14] RMedina J Llamas J Gomez-Garcıa-Bermejo E Zalama andM Segarra ldquoCrack detection in concrete tunnels using a Gaborfilter invariant to rotationrdquo Sensors vol 17 no 7 p 1670 2017
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Journal of Sensors 3
Point clouds ofmetro tunnel
Triangle mesh constructionCylinder fitting
Surface area calculation
Triangle mesh construction
Projection area calculation
Concrete spalling
Roughness descriptor
Outliers removal
Classification algorithm
Figure 2 Flowchart of the proposed method
(a) (b) (c)
Figure 3 Three types of outlier points (a) Drift points (b) Redundant points (c) Mixed points
areas thus the remaining patches are considered as belongingto concrete spalling This is given in Section 24
22 Outlier Points Removal The point cloud used for con-crete spalling detection is collected by a mobile three-dimensional laser scanning system that scans the tunnelsurface in the form of a section during movementThereforethe captured point cloud data is stored as multiple scan lineswhich not only contains the information of tunnel surface
but also captures the outlier points mainly originating fromcables lighting equipment pipelines and other facilitiesattached to the inner wall subway tracks and noise generatedby the scanner These outliers however inevitably affectthe identification of concrete spalling and thus should beremoved at the very beginning In our method the outlierpoints are roughly grouped into three categories accordingto spatial distribution namely drift points redundant pointsand mixed points as shown in Figure 3
4 Journal of Sensors
The drift points are caused by the random noise gen-erated by the laser scanner whose spatial distribution ischaracterized as being dispersed and far away from themain part of metro tunnel Thus the drift points can beeliminated by the clustering algorithm In addition theredundant points come from subway track and other groundparts in metro tunnel and the mixed points stem fromsome ancillary facilities However the two types of points areusually mixed with nonoutlier points of the tunnel surfacethus it is necessary to adopt an effective method to realizefiltering
Since the tunnel cross-section is designed as a standardcircle the captured point clouds present in the form of alarge number of circular scanning lines According to this wepropose a filtering algorithm to remove redundant points andmixed points by identifying points with large residual errorEach circular scan line is first fitted with a circle model bythe RANSAC (random sample consistency) [26] in whichthe circlersquos boundary and centre are obtained After that theresidual error can be calculated as the distance between eachpoint on the scan line and the circlersquos boundary Finally thethreshold of residual error is set and the points with residualerrors larger than the threshold can be seen as outliers andremoved
23 Roughness Descriptor Construction From the perspec-tive of topography surface roughness refers to the unevennessof the ground generally defined as the ratio of the surfacearea to the projected area for a unit It is usually used toreflect the high and low undulations on the terrain thephenomenon of which has a similar shape to the undulationson the tunnel surface Therefore in this paper the concept ofsurface roughness is introduced into the metro tunnel for therough areas recognition
231 Surface Area and Projected Area Calculation A rough-ness descriptor needs to be constructed to identify the roughareas on tunnel surface According to the definition of surfaceroughness (the ratio of surface area to the projected area for aunit) in our method the surface area of a unit is representedby a polygon area enclosed by the points of the first-orderneighbourhood around each point Similarly points on thetunnel surface are projected onto the cylindrical surface andthe polygon area of the first-order neighbourhood aroundeach corresponding projected point is taken as the projectedarea of a unit
Since the tunnel surface is nonplanar in order to calculatethe surface area or the projected area around each point itis necessary to construct a triangular network for the pointcloud of tunnel as well as the corresponding projected pointcloud of cylinder respectively To this end Poisson surfacereconstruction [27] is adopted in this subsection whichis an intuitive method for mesh construction with pointcloud and its normal vector serving as input componentswhile the output manifests as a three-dimensional gridFigure 4 shows the modelling results of a part of tunnelpoint cloud and the details of triangular mesh Given therough situation on tunnel surface we do not calculate the
Figure 4 Triangular mesh of tunnel surface
Figure 5 Triangular mesh of cylindrical surface
area of the polygon enclosed by the points of the first-orderneighbourhood but the sum of triangular areas of the first-order neighbourhood around each point as the surface areaof a unit
What is more is that for the sake of calculating theprojected area the corresponding relationship of pointsbetween tunnel and cylinder needs to be established accu-rately Lei You et al [28] proposed an algorithm for projectingtrunk point clouds onto a cylindrical surface in sections toreconstruct the trunk surface the theory of which is alsoapplicable to the tunnel surface Based on the algorithm thetunnel surface can be defined by two parameters The firstone is centreline (L) that is described by a series of centrepoints (c119894(119888119909 119888119910 119888119911) 119894 = 1 sdot sdot sdot 119899)The second parameter is thedesign diameter (d) of a metro tunnel Taking into accountthe coordinate system of tunnel point cloud Z axis is locatedat the vertical scanning plane with upward direction positiveand both 119883 and 119884 axes are located at the lateral scanningplane and perpendicular to each other which forms a right-handed coordinate system wherein the positive directionof X axis points to the mileage direction Thus for anypoint 119901(119901119909119901119910119901119911) on the tunnel surface its correspondingprojection point 1199011015840 on the cylindrical surface satisfies
10038171003817100381710038171003817(1199011015840 minus (119888119910 119888119911 119901119909)) times (1 0 0)10038171003817100381710038171003817 minus 119889 = 0 (1)
where times is the outer product of vector and representsthe modulus of the vector It is noted that Equation (1)will have two solutions taking the point close to 119901 as theprojection point 1199011015840 on the cylindrical surface Figure 5 showsthe modelling results of a part of tunnel point cloud afterprojection and the details of triangularmeshThen accordingto the coordinates of projection points the polygon areaenclosed by the points of the first-order neighbourhoodaround each projection point is calculated as the projectedarea of a unit
Journal of Sensors 5
T
AB
D
E
F
G
normal area
rough area
Figure 6 Sketch of rough area points
232 Roughness Descriptor of Tunnel Surface In this paperwe define the ratio of surface area to the projected area aroundeach point as the roughness descriptor After constructing thetriangular mesh for the tunnel point cloud it is necessaryto calculate the area of each triangle and then find the sumof areas of first-order neighbourhood triangles around eachpoint which is considered as the surface area of a unitSimilarly point clouds on the cylindrical surface obtainedfrom projection also need to generate a triangulated gridwhere the polygon area enclosed by the points of the first-order neighbourhood for each point is performed as theprojected area of a unit
Simulating a set of points for a microelement on thetunnel surface as shown in Figure 6 the black points repre-sent normal area while the yellow zone represents the rougharea It is assumed that the red point T in the picture isin the rough area and other points marked with A(1199091 1199101)B(1199092 1199102) and D(1199093 1199103) etc are the normal points around TThe area of triangle which ismade up of point T and the othertwo points around T can be calculated by Heronrsquos formula[29] and stored Taking the triangle 119879119860119861 as an examplelengths of the corresponding three sides are represented as 119905119886 and 119887 respectively thus the area of which can be calculatedby
119878119879119860119861 = radic119902 (119902 minus 119905) (119902 minus 119886) (119902 minus 119887) q = (119905 + 119886 + 119887)2 (2)
Therefore the sum of triangle areas of the first-order neigh-bourhood around point T can be expressed as follows
Finally the roughness threshold should be set accurately toextract the points in high rough areas on tunnel surface
24 Rough Area Classification The points in rough areas onthe surface of metro tunnel can be extracted by the roughnessdescriptor successfully which are composed of three maincategories namely concrete spalling patches bolt holes andsegment seams However the three types of points extractedbased on the roughness descriptor are mixed together sowe need to separate the points belonging to the concretespalling patches from the rough points In our methodtaking into account the irregularity of spalling patches wecannot directly identify them from rough areas Accordingto this the method of rough area classification is adoptedto accurately label patches belonging to segment seams andbolt holes so that the remaining patches are considered asbelonging to concrete spalling
For the seams between segments if the tunnel point cloudis unfolded the seam appears as a straight line Thus in thispaper themethod for seam recognition is to project the pointcloud of metro tunnel onto a plane and rasterize it into animage After that the Hough transform [30] is applied torecognize the lines so that the seams of tunnel segments canbe determined and furthermore eliminated from the pointcloud
For the bolt holes there is a fixed size we can establish astandard point cloud template of bolt holes The point cloudis then clustered and the degree of similarity between eachsmall clustered group and the template is compared based onthe similarity analysismethod to determinewhich small clus-ter belongs to bolt hole In order to obtain the clustered pointcloud of bolt hole the mean-shift clustering algorithm [31] isapplied to the remaining point cloud after seam eliminationincluding bolt holes and spalling patches so that the pointclouds can form many different small groups Meanwhilewe established the point cloud library of bolt hole to beregarded as a template for recognition The size of bolt holewe used to collect the point cloud library is about 20lowast14lowast18(cm) and 17lowast15lowast18 (cm) Based on the similarity comparisonbetween the template and the clustered small groups boltholes can be identified from the point clouds The specificimplementationmethod is as follows Firstly it is necessary tostandardize the position of template points and each clusteredsmall group by performing PCA transformation (Principal
6 Journal of Sensors
Small group points
PCA transformationTemplate points
ICP registration
Feature vector calculation
Similarity calculation
Bolt hole
Principal componentsof small group points
Principal componentsof template points
Vector 1 Vector 2
Small group pointsafter registration
Figure 7 Procedure of bolt hole recognition
Component Analysis) [32] on the two three-dimensionalpoint clouds so that the three main components of bothare obtained and taken as the new standardized coordinatesystems Secondly after the coordinate transformation eachsmall group is registered with the template by using the ICP(Iterative Closest Point) [33] registration algorithm to furtheradjust the clustered points so as to have a similar postureto the template as much as possible Thirdly calculate thefeature vector of point cloud whose method is proposed byXiaotong H et al [34] for the principal components of thetemplate and the registered small group points respectivelyand furthermore perform similarity comparison between thetwo feature vectors to distinguish bolt holes from roughpoints Generally speaking any small group of point cloudswith similarity score greater than the accurate thresholdcan be identified as a bolt hole The procedure of bolt holerecognition is shown in Figure 7
25 Detectable Spalling Analysis In this paper we define theratio of surface area to the projected area around each pointas the roughness descriptor and simultaneously a formulathereof has also been deduced in Section 232 What is moreis that it is necessary to analyse theminimum spalling patches
that can be extracted using this method in metro tunnelTherefore it is assumed that Figure 6 shows a microelementon the tunnel surface under the ideal conditions wherethe point spacing is represented by 119898 The black point isin the normal area whose depth is zero while the redpoint is in the spalling area and the depth is ℎ Accordingto the roughness formula the sum of areas of first-orderneighbourhood triangles around point T can be expressed as
Hence one can see that in the position where the spallingdoes not occur or the nonrough position h=0 that is the
Journal of Sensors 7
Figure 8A scan line of tunnel point cloudTheblack ring representsthe real position of a section on the tunnel inner wall and thedeviation of red points away from black ring is regarded as theprecision of point cloud (Δ)
value of roughness descriptor is 1 When h gt 0 the value ofroughness descriptor is greater than 1 It indicates that theposition is rough relative to the normal position and may bespalling
Therefore the detectability of concrete spalling basedon the roughness descriptor is determined by the spallingdepth ℎ and the point spacingm while these two parametersare mainly affected by the instrument accuracy and theset parameters of the MLS system used when collectingpoint cloud in metro tunnel namely range error Δ and theresolution of laser scanner as well as the running velocityof MLS system Firstly the range error Δ of laser scannerindicates the precision of the collected point cloud of metrotunnel Taking out a scan line of point cloud and expanding itinto a straight line as shown by red points in Figure 8 assumethat the black ring is the real position of a section on thetunnel inner wall while the deviation of captured points awayfrom the innerwall is regarded as the precision of point cloudwhich is represented byΔThus it can be seen that the spallingpatches will not be detected when the value of spalling depthℎ is less than Δ
The other factor that affects the detectability of spallingis the point spacing m including the vertical spacing andlongitudinal spacing The vertical spacing of point clouddepends on the resolution of the scanner When settingdifferent resolutions the number of scanning points on theone scan-line changes accordingly In addition since thefrequency of scanner is usually fixed the velocity of themobile laser scanning system determines the point spacingin the direction of the mileage commonly referred to asthe longitudinal spacing Taking a microelement on thesurface of tunnel as an example in Figure 9 the verticaland longitudinal spacing of point cloud are represented by1198981 and 1198982 respectively and the blue areas are used toindicate the spalling patches It follows that when the areaof spalling patches less than the product of vertical spacingand longitudinal spacing it cannot be detected either
Therefore when the depth ℎ and the area 119878119888 of a concretespalling patch satisfy the following formula (9) it can beextracted from the point cloud of tunnel surface whichcan also be used as a guideline to select optimal scanningparameters for MLS system
ℎ gt 119878119888 gt 1198981 lowast 1198982 (9)
m1
m2
Figure 9 A microelement on tunnel surface The red dots indicatethe points on a microelement of tunnel surface captured by scannerand the blue areas indicate the spalling patches
Figure 10 Mobile laser scanning system
3 Case Study
31 Data Collection of Metro Tunnel A section of a metrotunnel in Shanghai was selected as the experimental area witha total length of about 250 meters The mobile laser scanning(MLS) system is equippedwith a scanner of FAROFOCUS3DX330 for point cloud data collection in the tunnel as shownin Figure 10 the scanner of which has a scanning range of300∘ and working frequency of 100Hz In order not to affectthe routine operation of the subway the experimental dataacquisition was carried out between midnight and three inthe morning And the resolution of scanner is set to 14 sothe number of points in one circular scan line is about 9760and the vertical point spacing 1198981 is about 2mm Generallyduring the period of data acquisition to ensure the densityof point cloud the running velocity of MLS system on thesubway track is set to 05ms so the average point spacing1198982 of the collected point clouds in the mileage direction isabout 5mm The general information of the case area anddata collection is shown in Table 1
32 Experimental Results
321 Outlier Removal Result of Tunnel Point Cloud Thecaptured point cloud data by MLS system mainly containsthe information of the tunnel surface where it is also mixedwith some outlier points originating from the subway trackscables lighting equipment and other facilities which willinevitably have great interference on the concrete spalling
8 Journal of Sensors
Resid
ual e
rror
Point serial number
004
002
0
minus002
minus004
minus006
minus008
minus01
minus0120 500 1000 1500 2000 2500 3000 3500
(a)
(b) (c)
Figure 11 Outlier removal (a) Residual error curve of one scan line (b) before the outlier removal and (c) after the outlier removal
Table 1 General information of the case area and data collection
Parameters ValueLength of case tunnel 250 mRadius of the tunnel 275 mAverage velocity during data collection 05 msScanning distance 330 mScanning range 300∘
Working frequency 100 HzResolution 14Range error () 2 mmVertical point spacing (1198981) 2 mmLongitudinal point spacing (1198982) 5 mmPoint density 100000 ptsm2
Total points gt 500000000 pts
identification Thus according to the outlier points removalalgorithm introduced in Section 22 the residual error curvefor each circular scan line can be generated an example ofwhich is shown in Figure 11(a) Since the fluctuation range ofresidual errors is between plus and minus 001 hence it canbe seen that points with an error of less than negative 001 canbe considered outliers and then eliminated Figures 11(b) and11(c) show the point cloud of tunnel before and after outlierpoints removal respectively
Most of the outlier points with a certain distance fromthe tunnel surface can be removed using the residual errorfiltering algorithm However there are still a small fractionof points from the bottom of pipeline facility that clings tothe tunnel inner wall and cannot be completely eliminatedthrough thismethod causing them to eventually be identifiedas rough areas This part of points is usually presented in theform of a line so it can be identified and further removedtogether with ring seams through the algorithm of Houghtransformation later
322 Rough Area Extraction After removing the outlierpoints from original point cloud data of metro tunnel aroughness descriptor based method is applied to extract thepoints of rough areas on tunnel surface for the purpose offurther identifying the concrete spalling patches therefromThus in order to calculate the value of roughness descriptorfor each point triangular meshes are first constructed forboth the remaining points after outlier removal and thecorresponding projected points on the cylindrical surfacethereby obtaining the surface area and projected area ofthe first-order neighbourhood around each point Thenaccording to the definition of roughness descriptor the ratioof surface area to the projected area around each point iscalculated a histogram of which is also generated indicatingthe number of points corresponding to different roughnessvalues as shown in Figure 12
Journal of Sensors 9
times105
poin
t num
ber
roughness descriptor
2
18
16
14
12
1
08
06
04
02
009 1 11 12 13 14 15 16 17
Figure 12 Histogram of roughness descriptor
Figure 13 Roughness map of tunnel surface
According to the value of the roughness descriptor foreach point obtained we need to determine an accuratethreshold to extract the points belonging to rough areasTaking into account the scanner accuracy and the set param-eters of MLS system during operation in this experimentmainly relying on the range error and the point spacing119898 the expression of roughness descriptor can be furtherrepresented as follows
It can be seen from formula (10) that when the value ofthe depth ℎ for a certain point tends to zero roughness ratiois at a critical condition Therefore with the value of pointspacing119898 and range error Δ in this case study the thresholdof roughness descriptor has been obtained as about 105 andpoints with proportion greater than 105 are filtered as roughareas Expand the rough point cloud on the tunnel surfaceinto a plane as shown in Figure 13
It should be noted that we cannot thin the original pointcloud otherwise some points belonging to rough areas maybe missed which will affect the detection of spalling damageHowever when calculating the roughness value of each pointthe number of tunnel point clouds captured byMLS system isvery huge which will take a lot of time to find a polygon area
surrounded by its first-order neighbourhood points for eachpoint Accordingly BitMap and BloomFilter are adopted inthis part to improve performing efficiency of the algorithmBitmap is a compact data storage structure that allocates 1 bitof memory for each element in the collection which greatlyreduces the storage space required to process massive pointcloud data Based on this structure BloomFilter completesthe query of the first-order neighbourhood points for eachpoint and then we can calculate the area of polygon enclosedby them which greatly compresses the memory space andshortens the calculation time
323 Results of Concrete Spalling Detection The rough areasextracted on tunnel surface based on the method of rough-ness descriptor mainly contain three types of objects namelyconcrete spalling patches bolt holes and the seams betweensegments In order to identify the points belonging to spallingpatches we used the method of rough area classificationdescribed in Section 24 to separate bolt holes and the seg-ment seams from rough points so that the concrete spallingpatches can be remained
Firstly if the tunnel surface is unfolded into a planethe seam appears as a line Therefore the method of seamidentification is to project the rough point cloud extractedonto a plane and rasterize it into an image Then the Houghtransformation algorithm can be used to identify seamsbetween segments the result of which is shown in Figure 14As we can see from the figure seams including the transverseseams longitudinal seams and oblique seams have beenidentified successfully
Then the bolt holes are detected using the similarityanalysis method the separation result of which is shown inFigure 14 where the bolt holes are shown in blue blocksThe remaining patches belong to the concrete spalling areashown in red After detecting the tunnel section of 250m itwas found through statistics that the spalling patches mainlyoccurred in the subinterval between 175 and 200m so that inorder to express the detected spalling more clearly Figure 14shows the information of tunnel between 175 and 200mFinally we found seven concrete spalling patches in this case
10 Journal of Sensors
Table 2 Basic information of concrete spalling patches
No Mileage [m] Spalling patches Images Spalling area [m2] Spalling depth [m]
J1 176 00191 0112
J2 176 00276 0126
J3 177 00377 0195
J4 178 00340 0099
J5 177 00384 0129
J6 177 00681 0083
J7 193 00249 0091
study the basic information of which is shown in Table 2and the mileage position corresponding to the spalling is alsogiven at the same time
According to formula (9) the theoretical depth value andarea value of minimum spalling patch are 2mm and 10mm2respectively and the extracted results are indeed greater thanthe theoretical minimum
In this paper in order to examine the accuracy of thedetected results we conducted a jointmanual inspectionwith
the maintenance company on the seven detected spallingpatches that is each of the spalling was confirmed one by onein the tunnel It was found that each spalling did occur at thecorresponding position Therefore the false detection rate iszero and it is confirmed that the proposed concrete spallingdetection algorithm performs well In addition taking intoaccount the small size of the spalling and the large spacingbetween the points it does cause omission errors which havebeen analysed in Section 25 However these regions that
Journal of Sensors 11
Bolt holeConcrete spalling
Segment seam
7
6 5 4
32
1
Figure 14 Recognition result of bolt holes segment seams andconcrete spalling patches in tunnel subinterval of 175-200 m
could not be detected are very small and belong to the normalcategory which will not affect the performance and reliabilityof a tunnel
4 Conclusion
In this paper a new method used for concrete spallingdetection in metro tunnel from point cloud based on theroughness descriptor is proposed Firstly the point cloudacquired by mobile laser scanning system needs to eliminateoutlier points originating from ancillary facilities attachedto shield tunnel wall based on the residual error filteringalgorithmThen a roughness descriptor for the metro tunnelsurface is designed to extract the rough areas on the tunnelsurface including bolt holes segment seams and spallingpatches Finally rough area classification is performed onthe identified rough areas to accurately separate the segmentseams and bolt holes from rough areas so that the concretespalling patches are left A section of metro tunnel intervalabout 250m in Shanghai is selected to verify the validityof the proposed method and seven concrete spalling areasare detected which are identified as surface defects in metrotunnel This could be helpful for tunnel maintenance andoperation safety Compared with previous studies the con-cept of roughness descriptor is proposed to detect concretespalling which is suitable for not only flat concrete surfacesbut also nonplanar concrete surfaces and at the same timeoffer the guidance for optimal scanning parameter selection
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This research is supported by the National Science Foun-dation of China (no 41671451) the National Science andTechnologyMajor Program (2016YFB0502104) and the Fun-damental Research Funds for the Central Universities ofChina The authors would like to express appreciation tocolleagues in our laboratory for their valuable commentshelp
References
[1] T Asakura and Y Kojima ldquoTunnel maintenance in JapanrdquoTunnelling and Underground Space Technology vol 18 no 2-3pp 161ndash169 2003
[2] Y Yuan Y Bai and J Liu ldquoAssessment service state of tunnelstructurerdquo Tunnelling and Underground Space Technology vol27 no 1 pp 72ndash85 2012
[3] F Sandrone and V Labiouse ldquoIdentification and analysisof Swiss National Road tunnels pathologiesrdquo Tunnelling andUnderground Space Technology vol 26 no 2 pp 374ndash390 2011
[4] Portland Cement Association (PCA) Concrete slab surfacedefects Causes Prevention And Repair Portland CementSkokie IL USA 2001
[5] N Delatte S Chen N Maini et al ldquoApplication of non-destructive evaluation to subway tunnel systemsrdquo Transporta-tion Research Record vol 1845 no 3 pp 127ndash135 2003
[6] H Russell and J Gilmore ldquoInspection policy and proceduresfor rail transit tunnels and underground structuresrdquo TransitCooperative Research Program Synthesis of Transit Practice 1997
[7] Y Yuan X Jiang and Q Ai ldquoProbabilistic assessment forconcrete spalling in tunnel structuresrdquo ASCE-ASME Journalof Risk and Uncertainty in Engineering Systems Part A CivilEngineering vol 3 no 4 2017
[8] M-K Kim H Sohn and C-C Chang ldquoLocalization andquantification of concrete spalling defects using terrestrial laserscanningrdquo Journal of Computing in Civil Engineering vol 29 no6 2015
[9] Z Zhu and I Bilakis ldquoDetecting air pockets for architecturalconcrete quality assessment using visual sensingrdquo ElectronicJournal of Information Technology in Construction vol 13 pp86ndash102 2008
[10] American Concrete Institute ACI Manual of Concrete Inspec-tion ACI Committee 311 SP-2(07) Detroit Mich USA 2007
[11] A M Paterson G R Dowling and D A ChamberlainldquoBuilding inspection can computer vision helprdquo Automationin Construction vol 7 no 1 pp 13ndash20 1997
[12] B Guldur Erkal and J F Hajjar ldquoLaser-based surface damagedetection and quantification using predicted surface proper-tiesrdquo Automation in Construction vol 83 pp 285ndash302 2017
[13] T Dawood Z Zhu and T Zayed ldquoMachine vision-basedmodelfor spalling detection and quantification in subway networksrdquoAutomation in Construction vol 81 pp 149ndash160 2017
[14] RMedina J Llamas J Gomez-Garcıa-Bermejo E Zalama andM Segarra ldquoCrack detection in concrete tunnels using a Gaborfilter invariant to rotationrdquo Sensors vol 17 no 7 p 1670 2017
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
4 Journal of Sensors
The drift points are caused by the random noise gen-erated by the laser scanner whose spatial distribution ischaracterized as being dispersed and far away from themain part of metro tunnel Thus the drift points can beeliminated by the clustering algorithm In addition theredundant points come from subway track and other groundparts in metro tunnel and the mixed points stem fromsome ancillary facilities However the two types of points areusually mixed with nonoutlier points of the tunnel surfacethus it is necessary to adopt an effective method to realizefiltering
Since the tunnel cross-section is designed as a standardcircle the captured point clouds present in the form of alarge number of circular scanning lines According to this wepropose a filtering algorithm to remove redundant points andmixed points by identifying points with large residual errorEach circular scan line is first fitted with a circle model bythe RANSAC (random sample consistency) [26] in whichthe circlersquos boundary and centre are obtained After that theresidual error can be calculated as the distance between eachpoint on the scan line and the circlersquos boundary Finally thethreshold of residual error is set and the points with residualerrors larger than the threshold can be seen as outliers andremoved
23 Roughness Descriptor Construction From the perspec-tive of topography surface roughness refers to the unevennessof the ground generally defined as the ratio of the surfacearea to the projected area for a unit It is usually used toreflect the high and low undulations on the terrain thephenomenon of which has a similar shape to the undulationson the tunnel surface Therefore in this paper the concept ofsurface roughness is introduced into the metro tunnel for therough areas recognition
231 Surface Area and Projected Area Calculation A rough-ness descriptor needs to be constructed to identify the roughareas on tunnel surface According to the definition of surfaceroughness (the ratio of surface area to the projected area for aunit) in our method the surface area of a unit is representedby a polygon area enclosed by the points of the first-orderneighbourhood around each point Similarly points on thetunnel surface are projected onto the cylindrical surface andthe polygon area of the first-order neighbourhood aroundeach corresponding projected point is taken as the projectedarea of a unit
Since the tunnel surface is nonplanar in order to calculatethe surface area or the projected area around each point itis necessary to construct a triangular network for the pointcloud of tunnel as well as the corresponding projected pointcloud of cylinder respectively To this end Poisson surfacereconstruction [27] is adopted in this subsection whichis an intuitive method for mesh construction with pointcloud and its normal vector serving as input componentswhile the output manifests as a three-dimensional gridFigure 4 shows the modelling results of a part of tunnelpoint cloud and the details of triangular mesh Given therough situation on tunnel surface we do not calculate the
Figure 4 Triangular mesh of tunnel surface
Figure 5 Triangular mesh of cylindrical surface
area of the polygon enclosed by the points of the first-orderneighbourhood but the sum of triangular areas of the first-order neighbourhood around each point as the surface areaof a unit
What is more is that for the sake of calculating theprojected area the corresponding relationship of pointsbetween tunnel and cylinder needs to be established accu-rately Lei You et al [28] proposed an algorithm for projectingtrunk point clouds onto a cylindrical surface in sections toreconstruct the trunk surface the theory of which is alsoapplicable to the tunnel surface Based on the algorithm thetunnel surface can be defined by two parameters The firstone is centreline (L) that is described by a series of centrepoints (c119894(119888119909 119888119910 119888119911) 119894 = 1 sdot sdot sdot 119899)The second parameter is thedesign diameter (d) of a metro tunnel Taking into accountthe coordinate system of tunnel point cloud Z axis is locatedat the vertical scanning plane with upward direction positiveand both 119883 and 119884 axes are located at the lateral scanningplane and perpendicular to each other which forms a right-handed coordinate system wherein the positive directionof X axis points to the mileage direction Thus for anypoint 119901(119901119909119901119910119901119911) on the tunnel surface its correspondingprojection point 1199011015840 on the cylindrical surface satisfies
10038171003817100381710038171003817(1199011015840 minus (119888119910 119888119911 119901119909)) times (1 0 0)10038171003817100381710038171003817 minus 119889 = 0 (1)
where times is the outer product of vector and representsthe modulus of the vector It is noted that Equation (1)will have two solutions taking the point close to 119901 as theprojection point 1199011015840 on the cylindrical surface Figure 5 showsthe modelling results of a part of tunnel point cloud afterprojection and the details of triangularmeshThen accordingto the coordinates of projection points the polygon areaenclosed by the points of the first-order neighbourhoodaround each projection point is calculated as the projectedarea of a unit
Journal of Sensors 5
T
AB
D
E
F
G
normal area
rough area
Figure 6 Sketch of rough area points
232 Roughness Descriptor of Tunnel Surface In this paperwe define the ratio of surface area to the projected area aroundeach point as the roughness descriptor After constructing thetriangular mesh for the tunnel point cloud it is necessaryto calculate the area of each triangle and then find the sumof areas of first-order neighbourhood triangles around eachpoint which is considered as the surface area of a unitSimilarly point clouds on the cylindrical surface obtainedfrom projection also need to generate a triangulated gridwhere the polygon area enclosed by the points of the first-order neighbourhood for each point is performed as theprojected area of a unit
Simulating a set of points for a microelement on thetunnel surface as shown in Figure 6 the black points repre-sent normal area while the yellow zone represents the rougharea It is assumed that the red point T in the picture isin the rough area and other points marked with A(1199091 1199101)B(1199092 1199102) and D(1199093 1199103) etc are the normal points around TThe area of triangle which ismade up of point T and the othertwo points around T can be calculated by Heronrsquos formula[29] and stored Taking the triangle 119879119860119861 as an examplelengths of the corresponding three sides are represented as 119905119886 and 119887 respectively thus the area of which can be calculatedby
119878119879119860119861 = radic119902 (119902 minus 119905) (119902 minus 119886) (119902 minus 119887) q = (119905 + 119886 + 119887)2 (2)
Therefore the sum of triangle areas of the first-order neigh-bourhood around point T can be expressed as follows
Finally the roughness threshold should be set accurately toextract the points in high rough areas on tunnel surface
24 Rough Area Classification The points in rough areas onthe surface of metro tunnel can be extracted by the roughnessdescriptor successfully which are composed of three maincategories namely concrete spalling patches bolt holes andsegment seams However the three types of points extractedbased on the roughness descriptor are mixed together sowe need to separate the points belonging to the concretespalling patches from the rough points In our methodtaking into account the irregularity of spalling patches wecannot directly identify them from rough areas Accordingto this the method of rough area classification is adoptedto accurately label patches belonging to segment seams andbolt holes so that the remaining patches are considered asbelonging to concrete spalling
For the seams between segments if the tunnel point cloudis unfolded the seam appears as a straight line Thus in thispaper themethod for seam recognition is to project the pointcloud of metro tunnel onto a plane and rasterize it into animage After that the Hough transform [30] is applied torecognize the lines so that the seams of tunnel segments canbe determined and furthermore eliminated from the pointcloud
For the bolt holes there is a fixed size we can establish astandard point cloud template of bolt holes The point cloudis then clustered and the degree of similarity between eachsmall clustered group and the template is compared based onthe similarity analysismethod to determinewhich small clus-ter belongs to bolt hole In order to obtain the clustered pointcloud of bolt hole the mean-shift clustering algorithm [31] isapplied to the remaining point cloud after seam eliminationincluding bolt holes and spalling patches so that the pointclouds can form many different small groups Meanwhilewe established the point cloud library of bolt hole to beregarded as a template for recognition The size of bolt holewe used to collect the point cloud library is about 20lowast14lowast18(cm) and 17lowast15lowast18 (cm) Based on the similarity comparisonbetween the template and the clustered small groups boltholes can be identified from the point clouds The specificimplementationmethod is as follows Firstly it is necessary tostandardize the position of template points and each clusteredsmall group by performing PCA transformation (Principal
6 Journal of Sensors
Small group points
PCA transformationTemplate points
ICP registration
Feature vector calculation
Similarity calculation
Bolt hole
Principal componentsof small group points
Principal componentsof template points
Vector 1 Vector 2
Small group pointsafter registration
Figure 7 Procedure of bolt hole recognition
Component Analysis) [32] on the two three-dimensionalpoint clouds so that the three main components of bothare obtained and taken as the new standardized coordinatesystems Secondly after the coordinate transformation eachsmall group is registered with the template by using the ICP(Iterative Closest Point) [33] registration algorithm to furtheradjust the clustered points so as to have a similar postureto the template as much as possible Thirdly calculate thefeature vector of point cloud whose method is proposed byXiaotong H et al [34] for the principal components of thetemplate and the registered small group points respectivelyand furthermore perform similarity comparison between thetwo feature vectors to distinguish bolt holes from roughpoints Generally speaking any small group of point cloudswith similarity score greater than the accurate thresholdcan be identified as a bolt hole The procedure of bolt holerecognition is shown in Figure 7
25 Detectable Spalling Analysis In this paper we define theratio of surface area to the projected area around each pointas the roughness descriptor and simultaneously a formulathereof has also been deduced in Section 232 What is moreis that it is necessary to analyse theminimum spalling patches
that can be extracted using this method in metro tunnelTherefore it is assumed that Figure 6 shows a microelementon the tunnel surface under the ideal conditions wherethe point spacing is represented by 119898 The black point isin the normal area whose depth is zero while the redpoint is in the spalling area and the depth is ℎ Accordingto the roughness formula the sum of areas of first-orderneighbourhood triangles around point T can be expressed as
Hence one can see that in the position where the spallingdoes not occur or the nonrough position h=0 that is the
Journal of Sensors 7
Figure 8A scan line of tunnel point cloudTheblack ring representsthe real position of a section on the tunnel inner wall and thedeviation of red points away from black ring is regarded as theprecision of point cloud (Δ)
value of roughness descriptor is 1 When h gt 0 the value ofroughness descriptor is greater than 1 It indicates that theposition is rough relative to the normal position and may bespalling
Therefore the detectability of concrete spalling basedon the roughness descriptor is determined by the spallingdepth ℎ and the point spacingm while these two parametersare mainly affected by the instrument accuracy and theset parameters of the MLS system used when collectingpoint cloud in metro tunnel namely range error Δ and theresolution of laser scanner as well as the running velocityof MLS system Firstly the range error Δ of laser scannerindicates the precision of the collected point cloud of metrotunnel Taking out a scan line of point cloud and expanding itinto a straight line as shown by red points in Figure 8 assumethat the black ring is the real position of a section on thetunnel inner wall while the deviation of captured points awayfrom the innerwall is regarded as the precision of point cloudwhich is represented byΔThus it can be seen that the spallingpatches will not be detected when the value of spalling depthℎ is less than Δ
The other factor that affects the detectability of spallingis the point spacing m including the vertical spacing andlongitudinal spacing The vertical spacing of point clouddepends on the resolution of the scanner When settingdifferent resolutions the number of scanning points on theone scan-line changes accordingly In addition since thefrequency of scanner is usually fixed the velocity of themobile laser scanning system determines the point spacingin the direction of the mileage commonly referred to asthe longitudinal spacing Taking a microelement on thesurface of tunnel as an example in Figure 9 the verticaland longitudinal spacing of point cloud are represented by1198981 and 1198982 respectively and the blue areas are used toindicate the spalling patches It follows that when the areaof spalling patches less than the product of vertical spacingand longitudinal spacing it cannot be detected either
Therefore when the depth ℎ and the area 119878119888 of a concretespalling patch satisfy the following formula (9) it can beextracted from the point cloud of tunnel surface whichcan also be used as a guideline to select optimal scanningparameters for MLS system
ℎ gt 119878119888 gt 1198981 lowast 1198982 (9)
m1
m2
Figure 9 A microelement on tunnel surface The red dots indicatethe points on a microelement of tunnel surface captured by scannerand the blue areas indicate the spalling patches
Figure 10 Mobile laser scanning system
3 Case Study
31 Data Collection of Metro Tunnel A section of a metrotunnel in Shanghai was selected as the experimental area witha total length of about 250 meters The mobile laser scanning(MLS) system is equippedwith a scanner of FAROFOCUS3DX330 for point cloud data collection in the tunnel as shownin Figure 10 the scanner of which has a scanning range of300∘ and working frequency of 100Hz In order not to affectthe routine operation of the subway the experimental dataacquisition was carried out between midnight and three inthe morning And the resolution of scanner is set to 14 sothe number of points in one circular scan line is about 9760and the vertical point spacing 1198981 is about 2mm Generallyduring the period of data acquisition to ensure the densityof point cloud the running velocity of MLS system on thesubway track is set to 05ms so the average point spacing1198982 of the collected point clouds in the mileage direction isabout 5mm The general information of the case area anddata collection is shown in Table 1
32 Experimental Results
321 Outlier Removal Result of Tunnel Point Cloud Thecaptured point cloud data by MLS system mainly containsthe information of the tunnel surface where it is also mixedwith some outlier points originating from the subway trackscables lighting equipment and other facilities which willinevitably have great interference on the concrete spalling
8 Journal of Sensors
Resid
ual e
rror
Point serial number
004
002
0
minus002
minus004
minus006
minus008
minus01
minus0120 500 1000 1500 2000 2500 3000 3500
(a)
(b) (c)
Figure 11 Outlier removal (a) Residual error curve of one scan line (b) before the outlier removal and (c) after the outlier removal
Table 1 General information of the case area and data collection
Parameters ValueLength of case tunnel 250 mRadius of the tunnel 275 mAverage velocity during data collection 05 msScanning distance 330 mScanning range 300∘
Working frequency 100 HzResolution 14Range error () 2 mmVertical point spacing (1198981) 2 mmLongitudinal point spacing (1198982) 5 mmPoint density 100000 ptsm2
Total points gt 500000000 pts
identification Thus according to the outlier points removalalgorithm introduced in Section 22 the residual error curvefor each circular scan line can be generated an example ofwhich is shown in Figure 11(a) Since the fluctuation range ofresidual errors is between plus and minus 001 hence it canbe seen that points with an error of less than negative 001 canbe considered outliers and then eliminated Figures 11(b) and11(c) show the point cloud of tunnel before and after outlierpoints removal respectively
Most of the outlier points with a certain distance fromthe tunnel surface can be removed using the residual errorfiltering algorithm However there are still a small fractionof points from the bottom of pipeline facility that clings tothe tunnel inner wall and cannot be completely eliminatedthrough thismethod causing them to eventually be identifiedas rough areas This part of points is usually presented in theform of a line so it can be identified and further removedtogether with ring seams through the algorithm of Houghtransformation later
322 Rough Area Extraction After removing the outlierpoints from original point cloud data of metro tunnel aroughness descriptor based method is applied to extract thepoints of rough areas on tunnel surface for the purpose offurther identifying the concrete spalling patches therefromThus in order to calculate the value of roughness descriptorfor each point triangular meshes are first constructed forboth the remaining points after outlier removal and thecorresponding projected points on the cylindrical surfacethereby obtaining the surface area and projected area ofthe first-order neighbourhood around each point Thenaccording to the definition of roughness descriptor the ratioof surface area to the projected area around each point iscalculated a histogram of which is also generated indicatingthe number of points corresponding to different roughnessvalues as shown in Figure 12
Journal of Sensors 9
times105
poin
t num
ber
roughness descriptor
2
18
16
14
12
1
08
06
04
02
009 1 11 12 13 14 15 16 17
Figure 12 Histogram of roughness descriptor
Figure 13 Roughness map of tunnel surface
According to the value of the roughness descriptor foreach point obtained we need to determine an accuratethreshold to extract the points belonging to rough areasTaking into account the scanner accuracy and the set param-eters of MLS system during operation in this experimentmainly relying on the range error and the point spacing119898 the expression of roughness descriptor can be furtherrepresented as follows
It can be seen from formula (10) that when the value ofthe depth ℎ for a certain point tends to zero roughness ratiois at a critical condition Therefore with the value of pointspacing119898 and range error Δ in this case study the thresholdof roughness descriptor has been obtained as about 105 andpoints with proportion greater than 105 are filtered as roughareas Expand the rough point cloud on the tunnel surfaceinto a plane as shown in Figure 13
It should be noted that we cannot thin the original pointcloud otherwise some points belonging to rough areas maybe missed which will affect the detection of spalling damageHowever when calculating the roughness value of each pointthe number of tunnel point clouds captured byMLS system isvery huge which will take a lot of time to find a polygon area
surrounded by its first-order neighbourhood points for eachpoint Accordingly BitMap and BloomFilter are adopted inthis part to improve performing efficiency of the algorithmBitmap is a compact data storage structure that allocates 1 bitof memory for each element in the collection which greatlyreduces the storage space required to process massive pointcloud data Based on this structure BloomFilter completesthe query of the first-order neighbourhood points for eachpoint and then we can calculate the area of polygon enclosedby them which greatly compresses the memory space andshortens the calculation time
323 Results of Concrete Spalling Detection The rough areasextracted on tunnel surface based on the method of rough-ness descriptor mainly contain three types of objects namelyconcrete spalling patches bolt holes and the seams betweensegments In order to identify the points belonging to spallingpatches we used the method of rough area classificationdescribed in Section 24 to separate bolt holes and the seg-ment seams from rough points so that the concrete spallingpatches can be remained
Firstly if the tunnel surface is unfolded into a planethe seam appears as a line Therefore the method of seamidentification is to project the rough point cloud extractedonto a plane and rasterize it into an image Then the Houghtransformation algorithm can be used to identify seamsbetween segments the result of which is shown in Figure 14As we can see from the figure seams including the transverseseams longitudinal seams and oblique seams have beenidentified successfully
Then the bolt holes are detected using the similarityanalysis method the separation result of which is shown inFigure 14 where the bolt holes are shown in blue blocksThe remaining patches belong to the concrete spalling areashown in red After detecting the tunnel section of 250m itwas found through statistics that the spalling patches mainlyoccurred in the subinterval between 175 and 200m so that inorder to express the detected spalling more clearly Figure 14shows the information of tunnel between 175 and 200mFinally we found seven concrete spalling patches in this case
10 Journal of Sensors
Table 2 Basic information of concrete spalling patches
No Mileage [m] Spalling patches Images Spalling area [m2] Spalling depth [m]
J1 176 00191 0112
J2 176 00276 0126
J3 177 00377 0195
J4 178 00340 0099
J5 177 00384 0129
J6 177 00681 0083
J7 193 00249 0091
study the basic information of which is shown in Table 2and the mileage position corresponding to the spalling is alsogiven at the same time
According to formula (9) the theoretical depth value andarea value of minimum spalling patch are 2mm and 10mm2respectively and the extracted results are indeed greater thanthe theoretical minimum
In this paper in order to examine the accuracy of thedetected results we conducted a jointmanual inspectionwith
the maintenance company on the seven detected spallingpatches that is each of the spalling was confirmed one by onein the tunnel It was found that each spalling did occur at thecorresponding position Therefore the false detection rate iszero and it is confirmed that the proposed concrete spallingdetection algorithm performs well In addition taking intoaccount the small size of the spalling and the large spacingbetween the points it does cause omission errors which havebeen analysed in Section 25 However these regions that
Journal of Sensors 11
Bolt holeConcrete spalling
Segment seam
7
6 5 4
32
1
Figure 14 Recognition result of bolt holes segment seams andconcrete spalling patches in tunnel subinterval of 175-200 m
could not be detected are very small and belong to the normalcategory which will not affect the performance and reliabilityof a tunnel
4 Conclusion
In this paper a new method used for concrete spallingdetection in metro tunnel from point cloud based on theroughness descriptor is proposed Firstly the point cloudacquired by mobile laser scanning system needs to eliminateoutlier points originating from ancillary facilities attachedto shield tunnel wall based on the residual error filteringalgorithmThen a roughness descriptor for the metro tunnelsurface is designed to extract the rough areas on the tunnelsurface including bolt holes segment seams and spallingpatches Finally rough area classification is performed onthe identified rough areas to accurately separate the segmentseams and bolt holes from rough areas so that the concretespalling patches are left A section of metro tunnel intervalabout 250m in Shanghai is selected to verify the validityof the proposed method and seven concrete spalling areasare detected which are identified as surface defects in metrotunnel This could be helpful for tunnel maintenance andoperation safety Compared with previous studies the con-cept of roughness descriptor is proposed to detect concretespalling which is suitable for not only flat concrete surfacesbut also nonplanar concrete surfaces and at the same timeoffer the guidance for optimal scanning parameter selection
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This research is supported by the National Science Foun-dation of China (no 41671451) the National Science andTechnologyMajor Program (2016YFB0502104) and the Fun-damental Research Funds for the Central Universities ofChina The authors would like to express appreciation tocolleagues in our laboratory for their valuable commentshelp
References
[1] T Asakura and Y Kojima ldquoTunnel maintenance in JapanrdquoTunnelling and Underground Space Technology vol 18 no 2-3pp 161ndash169 2003
[2] Y Yuan Y Bai and J Liu ldquoAssessment service state of tunnelstructurerdquo Tunnelling and Underground Space Technology vol27 no 1 pp 72ndash85 2012
[3] F Sandrone and V Labiouse ldquoIdentification and analysisof Swiss National Road tunnels pathologiesrdquo Tunnelling andUnderground Space Technology vol 26 no 2 pp 374ndash390 2011
[4] Portland Cement Association (PCA) Concrete slab surfacedefects Causes Prevention And Repair Portland CementSkokie IL USA 2001
[5] N Delatte S Chen N Maini et al ldquoApplication of non-destructive evaluation to subway tunnel systemsrdquo Transporta-tion Research Record vol 1845 no 3 pp 127ndash135 2003
[6] H Russell and J Gilmore ldquoInspection policy and proceduresfor rail transit tunnels and underground structuresrdquo TransitCooperative Research Program Synthesis of Transit Practice 1997
[7] Y Yuan X Jiang and Q Ai ldquoProbabilistic assessment forconcrete spalling in tunnel structuresrdquo ASCE-ASME Journalof Risk and Uncertainty in Engineering Systems Part A CivilEngineering vol 3 no 4 2017
[8] M-K Kim H Sohn and C-C Chang ldquoLocalization andquantification of concrete spalling defects using terrestrial laserscanningrdquo Journal of Computing in Civil Engineering vol 29 no6 2015
[9] Z Zhu and I Bilakis ldquoDetecting air pockets for architecturalconcrete quality assessment using visual sensingrdquo ElectronicJournal of Information Technology in Construction vol 13 pp86ndash102 2008
[10] American Concrete Institute ACI Manual of Concrete Inspec-tion ACI Committee 311 SP-2(07) Detroit Mich USA 2007
[11] A M Paterson G R Dowling and D A ChamberlainldquoBuilding inspection can computer vision helprdquo Automationin Construction vol 7 no 1 pp 13ndash20 1997
[12] B Guldur Erkal and J F Hajjar ldquoLaser-based surface damagedetection and quantification using predicted surface proper-tiesrdquo Automation in Construction vol 83 pp 285ndash302 2017
[13] T Dawood Z Zhu and T Zayed ldquoMachine vision-basedmodelfor spalling detection and quantification in subway networksrdquoAutomation in Construction vol 81 pp 149ndash160 2017
[14] RMedina J Llamas J Gomez-Garcıa-Bermejo E Zalama andM Segarra ldquoCrack detection in concrete tunnels using a Gaborfilter invariant to rotationrdquo Sensors vol 17 no 7 p 1670 2017
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Journal of Sensors 5
T
AB
D
E
F
G
normal area
rough area
Figure 6 Sketch of rough area points
232 Roughness Descriptor of Tunnel Surface In this paperwe define the ratio of surface area to the projected area aroundeach point as the roughness descriptor After constructing thetriangular mesh for the tunnel point cloud it is necessaryto calculate the area of each triangle and then find the sumof areas of first-order neighbourhood triangles around eachpoint which is considered as the surface area of a unitSimilarly point clouds on the cylindrical surface obtainedfrom projection also need to generate a triangulated gridwhere the polygon area enclosed by the points of the first-order neighbourhood for each point is performed as theprojected area of a unit
Simulating a set of points for a microelement on thetunnel surface as shown in Figure 6 the black points repre-sent normal area while the yellow zone represents the rougharea It is assumed that the red point T in the picture isin the rough area and other points marked with A(1199091 1199101)B(1199092 1199102) and D(1199093 1199103) etc are the normal points around TThe area of triangle which ismade up of point T and the othertwo points around T can be calculated by Heronrsquos formula[29] and stored Taking the triangle 119879119860119861 as an examplelengths of the corresponding three sides are represented as 119905119886 and 119887 respectively thus the area of which can be calculatedby
119878119879119860119861 = radic119902 (119902 minus 119905) (119902 minus 119886) (119902 minus 119887) q = (119905 + 119886 + 119887)2 (2)
Therefore the sum of triangle areas of the first-order neigh-bourhood around point T can be expressed as follows
Finally the roughness threshold should be set accurately toextract the points in high rough areas on tunnel surface
24 Rough Area Classification The points in rough areas onthe surface of metro tunnel can be extracted by the roughnessdescriptor successfully which are composed of three maincategories namely concrete spalling patches bolt holes andsegment seams However the three types of points extractedbased on the roughness descriptor are mixed together sowe need to separate the points belonging to the concretespalling patches from the rough points In our methodtaking into account the irregularity of spalling patches wecannot directly identify them from rough areas Accordingto this the method of rough area classification is adoptedto accurately label patches belonging to segment seams andbolt holes so that the remaining patches are considered asbelonging to concrete spalling
For the seams between segments if the tunnel point cloudis unfolded the seam appears as a straight line Thus in thispaper themethod for seam recognition is to project the pointcloud of metro tunnel onto a plane and rasterize it into animage After that the Hough transform [30] is applied torecognize the lines so that the seams of tunnel segments canbe determined and furthermore eliminated from the pointcloud
For the bolt holes there is a fixed size we can establish astandard point cloud template of bolt holes The point cloudis then clustered and the degree of similarity between eachsmall clustered group and the template is compared based onthe similarity analysismethod to determinewhich small clus-ter belongs to bolt hole In order to obtain the clustered pointcloud of bolt hole the mean-shift clustering algorithm [31] isapplied to the remaining point cloud after seam eliminationincluding bolt holes and spalling patches so that the pointclouds can form many different small groups Meanwhilewe established the point cloud library of bolt hole to beregarded as a template for recognition The size of bolt holewe used to collect the point cloud library is about 20lowast14lowast18(cm) and 17lowast15lowast18 (cm) Based on the similarity comparisonbetween the template and the clustered small groups boltholes can be identified from the point clouds The specificimplementationmethod is as follows Firstly it is necessary tostandardize the position of template points and each clusteredsmall group by performing PCA transformation (Principal
6 Journal of Sensors
Small group points
PCA transformationTemplate points
ICP registration
Feature vector calculation
Similarity calculation
Bolt hole
Principal componentsof small group points
Principal componentsof template points
Vector 1 Vector 2
Small group pointsafter registration
Figure 7 Procedure of bolt hole recognition
Component Analysis) [32] on the two three-dimensionalpoint clouds so that the three main components of bothare obtained and taken as the new standardized coordinatesystems Secondly after the coordinate transformation eachsmall group is registered with the template by using the ICP(Iterative Closest Point) [33] registration algorithm to furtheradjust the clustered points so as to have a similar postureto the template as much as possible Thirdly calculate thefeature vector of point cloud whose method is proposed byXiaotong H et al [34] for the principal components of thetemplate and the registered small group points respectivelyand furthermore perform similarity comparison between thetwo feature vectors to distinguish bolt holes from roughpoints Generally speaking any small group of point cloudswith similarity score greater than the accurate thresholdcan be identified as a bolt hole The procedure of bolt holerecognition is shown in Figure 7
25 Detectable Spalling Analysis In this paper we define theratio of surface area to the projected area around each pointas the roughness descriptor and simultaneously a formulathereof has also been deduced in Section 232 What is moreis that it is necessary to analyse theminimum spalling patches
that can be extracted using this method in metro tunnelTherefore it is assumed that Figure 6 shows a microelementon the tunnel surface under the ideal conditions wherethe point spacing is represented by 119898 The black point isin the normal area whose depth is zero while the redpoint is in the spalling area and the depth is ℎ Accordingto the roughness formula the sum of areas of first-orderneighbourhood triangles around point T can be expressed as
Hence one can see that in the position where the spallingdoes not occur or the nonrough position h=0 that is the
Journal of Sensors 7
Figure 8A scan line of tunnel point cloudTheblack ring representsthe real position of a section on the tunnel inner wall and thedeviation of red points away from black ring is regarded as theprecision of point cloud (Δ)
value of roughness descriptor is 1 When h gt 0 the value ofroughness descriptor is greater than 1 It indicates that theposition is rough relative to the normal position and may bespalling
Therefore the detectability of concrete spalling basedon the roughness descriptor is determined by the spallingdepth ℎ and the point spacingm while these two parametersare mainly affected by the instrument accuracy and theset parameters of the MLS system used when collectingpoint cloud in metro tunnel namely range error Δ and theresolution of laser scanner as well as the running velocityof MLS system Firstly the range error Δ of laser scannerindicates the precision of the collected point cloud of metrotunnel Taking out a scan line of point cloud and expanding itinto a straight line as shown by red points in Figure 8 assumethat the black ring is the real position of a section on thetunnel inner wall while the deviation of captured points awayfrom the innerwall is regarded as the precision of point cloudwhich is represented byΔThus it can be seen that the spallingpatches will not be detected when the value of spalling depthℎ is less than Δ
The other factor that affects the detectability of spallingis the point spacing m including the vertical spacing andlongitudinal spacing The vertical spacing of point clouddepends on the resolution of the scanner When settingdifferent resolutions the number of scanning points on theone scan-line changes accordingly In addition since thefrequency of scanner is usually fixed the velocity of themobile laser scanning system determines the point spacingin the direction of the mileage commonly referred to asthe longitudinal spacing Taking a microelement on thesurface of tunnel as an example in Figure 9 the verticaland longitudinal spacing of point cloud are represented by1198981 and 1198982 respectively and the blue areas are used toindicate the spalling patches It follows that when the areaof spalling patches less than the product of vertical spacingand longitudinal spacing it cannot be detected either
Therefore when the depth ℎ and the area 119878119888 of a concretespalling patch satisfy the following formula (9) it can beextracted from the point cloud of tunnel surface whichcan also be used as a guideline to select optimal scanningparameters for MLS system
ℎ gt 119878119888 gt 1198981 lowast 1198982 (9)
m1
m2
Figure 9 A microelement on tunnel surface The red dots indicatethe points on a microelement of tunnel surface captured by scannerand the blue areas indicate the spalling patches
Figure 10 Mobile laser scanning system
3 Case Study
31 Data Collection of Metro Tunnel A section of a metrotunnel in Shanghai was selected as the experimental area witha total length of about 250 meters The mobile laser scanning(MLS) system is equippedwith a scanner of FAROFOCUS3DX330 for point cloud data collection in the tunnel as shownin Figure 10 the scanner of which has a scanning range of300∘ and working frequency of 100Hz In order not to affectthe routine operation of the subway the experimental dataacquisition was carried out between midnight and three inthe morning And the resolution of scanner is set to 14 sothe number of points in one circular scan line is about 9760and the vertical point spacing 1198981 is about 2mm Generallyduring the period of data acquisition to ensure the densityof point cloud the running velocity of MLS system on thesubway track is set to 05ms so the average point spacing1198982 of the collected point clouds in the mileage direction isabout 5mm The general information of the case area anddata collection is shown in Table 1
32 Experimental Results
321 Outlier Removal Result of Tunnel Point Cloud Thecaptured point cloud data by MLS system mainly containsthe information of the tunnel surface where it is also mixedwith some outlier points originating from the subway trackscables lighting equipment and other facilities which willinevitably have great interference on the concrete spalling
8 Journal of Sensors
Resid
ual e
rror
Point serial number
004
002
0
minus002
minus004
minus006
minus008
minus01
minus0120 500 1000 1500 2000 2500 3000 3500
(a)
(b) (c)
Figure 11 Outlier removal (a) Residual error curve of one scan line (b) before the outlier removal and (c) after the outlier removal
Table 1 General information of the case area and data collection
Parameters ValueLength of case tunnel 250 mRadius of the tunnel 275 mAverage velocity during data collection 05 msScanning distance 330 mScanning range 300∘
Working frequency 100 HzResolution 14Range error () 2 mmVertical point spacing (1198981) 2 mmLongitudinal point spacing (1198982) 5 mmPoint density 100000 ptsm2
Total points gt 500000000 pts
identification Thus according to the outlier points removalalgorithm introduced in Section 22 the residual error curvefor each circular scan line can be generated an example ofwhich is shown in Figure 11(a) Since the fluctuation range ofresidual errors is between plus and minus 001 hence it canbe seen that points with an error of less than negative 001 canbe considered outliers and then eliminated Figures 11(b) and11(c) show the point cloud of tunnel before and after outlierpoints removal respectively
Most of the outlier points with a certain distance fromthe tunnel surface can be removed using the residual errorfiltering algorithm However there are still a small fractionof points from the bottom of pipeline facility that clings tothe tunnel inner wall and cannot be completely eliminatedthrough thismethod causing them to eventually be identifiedas rough areas This part of points is usually presented in theform of a line so it can be identified and further removedtogether with ring seams through the algorithm of Houghtransformation later
322 Rough Area Extraction After removing the outlierpoints from original point cloud data of metro tunnel aroughness descriptor based method is applied to extract thepoints of rough areas on tunnel surface for the purpose offurther identifying the concrete spalling patches therefromThus in order to calculate the value of roughness descriptorfor each point triangular meshes are first constructed forboth the remaining points after outlier removal and thecorresponding projected points on the cylindrical surfacethereby obtaining the surface area and projected area ofthe first-order neighbourhood around each point Thenaccording to the definition of roughness descriptor the ratioof surface area to the projected area around each point iscalculated a histogram of which is also generated indicatingthe number of points corresponding to different roughnessvalues as shown in Figure 12
Journal of Sensors 9
times105
poin
t num
ber
roughness descriptor
2
18
16
14
12
1
08
06
04
02
009 1 11 12 13 14 15 16 17
Figure 12 Histogram of roughness descriptor
Figure 13 Roughness map of tunnel surface
According to the value of the roughness descriptor foreach point obtained we need to determine an accuratethreshold to extract the points belonging to rough areasTaking into account the scanner accuracy and the set param-eters of MLS system during operation in this experimentmainly relying on the range error and the point spacing119898 the expression of roughness descriptor can be furtherrepresented as follows
It can be seen from formula (10) that when the value ofthe depth ℎ for a certain point tends to zero roughness ratiois at a critical condition Therefore with the value of pointspacing119898 and range error Δ in this case study the thresholdof roughness descriptor has been obtained as about 105 andpoints with proportion greater than 105 are filtered as roughareas Expand the rough point cloud on the tunnel surfaceinto a plane as shown in Figure 13
It should be noted that we cannot thin the original pointcloud otherwise some points belonging to rough areas maybe missed which will affect the detection of spalling damageHowever when calculating the roughness value of each pointthe number of tunnel point clouds captured byMLS system isvery huge which will take a lot of time to find a polygon area
surrounded by its first-order neighbourhood points for eachpoint Accordingly BitMap and BloomFilter are adopted inthis part to improve performing efficiency of the algorithmBitmap is a compact data storage structure that allocates 1 bitof memory for each element in the collection which greatlyreduces the storage space required to process massive pointcloud data Based on this structure BloomFilter completesthe query of the first-order neighbourhood points for eachpoint and then we can calculate the area of polygon enclosedby them which greatly compresses the memory space andshortens the calculation time
323 Results of Concrete Spalling Detection The rough areasextracted on tunnel surface based on the method of rough-ness descriptor mainly contain three types of objects namelyconcrete spalling patches bolt holes and the seams betweensegments In order to identify the points belonging to spallingpatches we used the method of rough area classificationdescribed in Section 24 to separate bolt holes and the seg-ment seams from rough points so that the concrete spallingpatches can be remained
Firstly if the tunnel surface is unfolded into a planethe seam appears as a line Therefore the method of seamidentification is to project the rough point cloud extractedonto a plane and rasterize it into an image Then the Houghtransformation algorithm can be used to identify seamsbetween segments the result of which is shown in Figure 14As we can see from the figure seams including the transverseseams longitudinal seams and oblique seams have beenidentified successfully
Then the bolt holes are detected using the similarityanalysis method the separation result of which is shown inFigure 14 where the bolt holes are shown in blue blocksThe remaining patches belong to the concrete spalling areashown in red After detecting the tunnel section of 250m itwas found through statistics that the spalling patches mainlyoccurred in the subinterval between 175 and 200m so that inorder to express the detected spalling more clearly Figure 14shows the information of tunnel between 175 and 200mFinally we found seven concrete spalling patches in this case
10 Journal of Sensors
Table 2 Basic information of concrete spalling patches
No Mileage [m] Spalling patches Images Spalling area [m2] Spalling depth [m]
J1 176 00191 0112
J2 176 00276 0126
J3 177 00377 0195
J4 178 00340 0099
J5 177 00384 0129
J6 177 00681 0083
J7 193 00249 0091
study the basic information of which is shown in Table 2and the mileage position corresponding to the spalling is alsogiven at the same time
According to formula (9) the theoretical depth value andarea value of minimum spalling patch are 2mm and 10mm2respectively and the extracted results are indeed greater thanthe theoretical minimum
In this paper in order to examine the accuracy of thedetected results we conducted a jointmanual inspectionwith
the maintenance company on the seven detected spallingpatches that is each of the spalling was confirmed one by onein the tunnel It was found that each spalling did occur at thecorresponding position Therefore the false detection rate iszero and it is confirmed that the proposed concrete spallingdetection algorithm performs well In addition taking intoaccount the small size of the spalling and the large spacingbetween the points it does cause omission errors which havebeen analysed in Section 25 However these regions that
Journal of Sensors 11
Bolt holeConcrete spalling
Segment seam
7
6 5 4
32
1
Figure 14 Recognition result of bolt holes segment seams andconcrete spalling patches in tunnel subinterval of 175-200 m
could not be detected are very small and belong to the normalcategory which will not affect the performance and reliabilityof a tunnel
4 Conclusion
In this paper a new method used for concrete spallingdetection in metro tunnel from point cloud based on theroughness descriptor is proposed Firstly the point cloudacquired by mobile laser scanning system needs to eliminateoutlier points originating from ancillary facilities attachedto shield tunnel wall based on the residual error filteringalgorithmThen a roughness descriptor for the metro tunnelsurface is designed to extract the rough areas on the tunnelsurface including bolt holes segment seams and spallingpatches Finally rough area classification is performed onthe identified rough areas to accurately separate the segmentseams and bolt holes from rough areas so that the concretespalling patches are left A section of metro tunnel intervalabout 250m in Shanghai is selected to verify the validityof the proposed method and seven concrete spalling areasare detected which are identified as surface defects in metrotunnel This could be helpful for tunnel maintenance andoperation safety Compared with previous studies the con-cept of roughness descriptor is proposed to detect concretespalling which is suitable for not only flat concrete surfacesbut also nonplanar concrete surfaces and at the same timeoffer the guidance for optimal scanning parameter selection
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This research is supported by the National Science Foun-dation of China (no 41671451) the National Science andTechnologyMajor Program (2016YFB0502104) and the Fun-damental Research Funds for the Central Universities ofChina The authors would like to express appreciation tocolleagues in our laboratory for their valuable commentshelp
References
[1] T Asakura and Y Kojima ldquoTunnel maintenance in JapanrdquoTunnelling and Underground Space Technology vol 18 no 2-3pp 161ndash169 2003
[2] Y Yuan Y Bai and J Liu ldquoAssessment service state of tunnelstructurerdquo Tunnelling and Underground Space Technology vol27 no 1 pp 72ndash85 2012
[3] F Sandrone and V Labiouse ldquoIdentification and analysisof Swiss National Road tunnels pathologiesrdquo Tunnelling andUnderground Space Technology vol 26 no 2 pp 374ndash390 2011
[4] Portland Cement Association (PCA) Concrete slab surfacedefects Causes Prevention And Repair Portland CementSkokie IL USA 2001
[5] N Delatte S Chen N Maini et al ldquoApplication of non-destructive evaluation to subway tunnel systemsrdquo Transporta-tion Research Record vol 1845 no 3 pp 127ndash135 2003
[6] H Russell and J Gilmore ldquoInspection policy and proceduresfor rail transit tunnels and underground structuresrdquo TransitCooperative Research Program Synthesis of Transit Practice 1997
[7] Y Yuan X Jiang and Q Ai ldquoProbabilistic assessment forconcrete spalling in tunnel structuresrdquo ASCE-ASME Journalof Risk and Uncertainty in Engineering Systems Part A CivilEngineering vol 3 no 4 2017
[8] M-K Kim H Sohn and C-C Chang ldquoLocalization andquantification of concrete spalling defects using terrestrial laserscanningrdquo Journal of Computing in Civil Engineering vol 29 no6 2015
[9] Z Zhu and I Bilakis ldquoDetecting air pockets for architecturalconcrete quality assessment using visual sensingrdquo ElectronicJournal of Information Technology in Construction vol 13 pp86ndash102 2008
[10] American Concrete Institute ACI Manual of Concrete Inspec-tion ACI Committee 311 SP-2(07) Detroit Mich USA 2007
[11] A M Paterson G R Dowling and D A ChamberlainldquoBuilding inspection can computer vision helprdquo Automationin Construction vol 7 no 1 pp 13ndash20 1997
[12] B Guldur Erkal and J F Hajjar ldquoLaser-based surface damagedetection and quantification using predicted surface proper-tiesrdquo Automation in Construction vol 83 pp 285ndash302 2017
[13] T Dawood Z Zhu and T Zayed ldquoMachine vision-basedmodelfor spalling detection and quantification in subway networksrdquoAutomation in Construction vol 81 pp 149ndash160 2017
[14] RMedina J Llamas J Gomez-Garcıa-Bermejo E Zalama andM Segarra ldquoCrack detection in concrete tunnels using a Gaborfilter invariant to rotationrdquo Sensors vol 17 no 7 p 1670 2017
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
6 Journal of Sensors
Small group points
PCA transformationTemplate points
ICP registration
Feature vector calculation
Similarity calculation
Bolt hole
Principal componentsof small group points
Principal componentsof template points
Vector 1 Vector 2
Small group pointsafter registration
Figure 7 Procedure of bolt hole recognition
Component Analysis) [32] on the two three-dimensionalpoint clouds so that the three main components of bothare obtained and taken as the new standardized coordinatesystems Secondly after the coordinate transformation eachsmall group is registered with the template by using the ICP(Iterative Closest Point) [33] registration algorithm to furtheradjust the clustered points so as to have a similar postureto the template as much as possible Thirdly calculate thefeature vector of point cloud whose method is proposed byXiaotong H et al [34] for the principal components of thetemplate and the registered small group points respectivelyand furthermore perform similarity comparison between thetwo feature vectors to distinguish bolt holes from roughpoints Generally speaking any small group of point cloudswith similarity score greater than the accurate thresholdcan be identified as a bolt hole The procedure of bolt holerecognition is shown in Figure 7
25 Detectable Spalling Analysis In this paper we define theratio of surface area to the projected area around each pointas the roughness descriptor and simultaneously a formulathereof has also been deduced in Section 232 What is moreis that it is necessary to analyse theminimum spalling patches
that can be extracted using this method in metro tunnelTherefore it is assumed that Figure 6 shows a microelementon the tunnel surface under the ideal conditions wherethe point spacing is represented by 119898 The black point isin the normal area whose depth is zero while the redpoint is in the spalling area and the depth is ℎ Accordingto the roughness formula the sum of areas of first-orderneighbourhood triangles around point T can be expressed as
Hence one can see that in the position where the spallingdoes not occur or the nonrough position h=0 that is the
Journal of Sensors 7
Figure 8A scan line of tunnel point cloudTheblack ring representsthe real position of a section on the tunnel inner wall and thedeviation of red points away from black ring is regarded as theprecision of point cloud (Δ)
value of roughness descriptor is 1 When h gt 0 the value ofroughness descriptor is greater than 1 It indicates that theposition is rough relative to the normal position and may bespalling
Therefore the detectability of concrete spalling basedon the roughness descriptor is determined by the spallingdepth ℎ and the point spacingm while these two parametersare mainly affected by the instrument accuracy and theset parameters of the MLS system used when collectingpoint cloud in metro tunnel namely range error Δ and theresolution of laser scanner as well as the running velocityof MLS system Firstly the range error Δ of laser scannerindicates the precision of the collected point cloud of metrotunnel Taking out a scan line of point cloud and expanding itinto a straight line as shown by red points in Figure 8 assumethat the black ring is the real position of a section on thetunnel inner wall while the deviation of captured points awayfrom the innerwall is regarded as the precision of point cloudwhich is represented byΔThus it can be seen that the spallingpatches will not be detected when the value of spalling depthℎ is less than Δ
The other factor that affects the detectability of spallingis the point spacing m including the vertical spacing andlongitudinal spacing The vertical spacing of point clouddepends on the resolution of the scanner When settingdifferent resolutions the number of scanning points on theone scan-line changes accordingly In addition since thefrequency of scanner is usually fixed the velocity of themobile laser scanning system determines the point spacingin the direction of the mileage commonly referred to asthe longitudinal spacing Taking a microelement on thesurface of tunnel as an example in Figure 9 the verticaland longitudinal spacing of point cloud are represented by1198981 and 1198982 respectively and the blue areas are used toindicate the spalling patches It follows that when the areaof spalling patches less than the product of vertical spacingand longitudinal spacing it cannot be detected either
Therefore when the depth ℎ and the area 119878119888 of a concretespalling patch satisfy the following formula (9) it can beextracted from the point cloud of tunnel surface whichcan also be used as a guideline to select optimal scanningparameters for MLS system
ℎ gt 119878119888 gt 1198981 lowast 1198982 (9)
m1
m2
Figure 9 A microelement on tunnel surface The red dots indicatethe points on a microelement of tunnel surface captured by scannerand the blue areas indicate the spalling patches
Figure 10 Mobile laser scanning system
3 Case Study
31 Data Collection of Metro Tunnel A section of a metrotunnel in Shanghai was selected as the experimental area witha total length of about 250 meters The mobile laser scanning(MLS) system is equippedwith a scanner of FAROFOCUS3DX330 for point cloud data collection in the tunnel as shownin Figure 10 the scanner of which has a scanning range of300∘ and working frequency of 100Hz In order not to affectthe routine operation of the subway the experimental dataacquisition was carried out between midnight and three inthe morning And the resolution of scanner is set to 14 sothe number of points in one circular scan line is about 9760and the vertical point spacing 1198981 is about 2mm Generallyduring the period of data acquisition to ensure the densityof point cloud the running velocity of MLS system on thesubway track is set to 05ms so the average point spacing1198982 of the collected point clouds in the mileage direction isabout 5mm The general information of the case area anddata collection is shown in Table 1
32 Experimental Results
321 Outlier Removal Result of Tunnel Point Cloud Thecaptured point cloud data by MLS system mainly containsthe information of the tunnel surface where it is also mixedwith some outlier points originating from the subway trackscables lighting equipment and other facilities which willinevitably have great interference on the concrete spalling
8 Journal of Sensors
Resid
ual e
rror
Point serial number
004
002
0
minus002
minus004
minus006
minus008
minus01
minus0120 500 1000 1500 2000 2500 3000 3500
(a)
(b) (c)
Figure 11 Outlier removal (a) Residual error curve of one scan line (b) before the outlier removal and (c) after the outlier removal
Table 1 General information of the case area and data collection
Parameters ValueLength of case tunnel 250 mRadius of the tunnel 275 mAverage velocity during data collection 05 msScanning distance 330 mScanning range 300∘
Working frequency 100 HzResolution 14Range error () 2 mmVertical point spacing (1198981) 2 mmLongitudinal point spacing (1198982) 5 mmPoint density 100000 ptsm2
Total points gt 500000000 pts
identification Thus according to the outlier points removalalgorithm introduced in Section 22 the residual error curvefor each circular scan line can be generated an example ofwhich is shown in Figure 11(a) Since the fluctuation range ofresidual errors is between plus and minus 001 hence it canbe seen that points with an error of less than negative 001 canbe considered outliers and then eliminated Figures 11(b) and11(c) show the point cloud of tunnel before and after outlierpoints removal respectively
Most of the outlier points with a certain distance fromthe tunnel surface can be removed using the residual errorfiltering algorithm However there are still a small fractionof points from the bottom of pipeline facility that clings tothe tunnel inner wall and cannot be completely eliminatedthrough thismethod causing them to eventually be identifiedas rough areas This part of points is usually presented in theform of a line so it can be identified and further removedtogether with ring seams through the algorithm of Houghtransformation later
322 Rough Area Extraction After removing the outlierpoints from original point cloud data of metro tunnel aroughness descriptor based method is applied to extract thepoints of rough areas on tunnel surface for the purpose offurther identifying the concrete spalling patches therefromThus in order to calculate the value of roughness descriptorfor each point triangular meshes are first constructed forboth the remaining points after outlier removal and thecorresponding projected points on the cylindrical surfacethereby obtaining the surface area and projected area ofthe first-order neighbourhood around each point Thenaccording to the definition of roughness descriptor the ratioof surface area to the projected area around each point iscalculated a histogram of which is also generated indicatingthe number of points corresponding to different roughnessvalues as shown in Figure 12
Journal of Sensors 9
times105
poin
t num
ber
roughness descriptor
2
18
16
14
12
1
08
06
04
02
009 1 11 12 13 14 15 16 17
Figure 12 Histogram of roughness descriptor
Figure 13 Roughness map of tunnel surface
According to the value of the roughness descriptor foreach point obtained we need to determine an accuratethreshold to extract the points belonging to rough areasTaking into account the scanner accuracy and the set param-eters of MLS system during operation in this experimentmainly relying on the range error and the point spacing119898 the expression of roughness descriptor can be furtherrepresented as follows
It can be seen from formula (10) that when the value ofthe depth ℎ for a certain point tends to zero roughness ratiois at a critical condition Therefore with the value of pointspacing119898 and range error Δ in this case study the thresholdof roughness descriptor has been obtained as about 105 andpoints with proportion greater than 105 are filtered as roughareas Expand the rough point cloud on the tunnel surfaceinto a plane as shown in Figure 13
It should be noted that we cannot thin the original pointcloud otherwise some points belonging to rough areas maybe missed which will affect the detection of spalling damageHowever when calculating the roughness value of each pointthe number of tunnel point clouds captured byMLS system isvery huge which will take a lot of time to find a polygon area
surrounded by its first-order neighbourhood points for eachpoint Accordingly BitMap and BloomFilter are adopted inthis part to improve performing efficiency of the algorithmBitmap is a compact data storage structure that allocates 1 bitof memory for each element in the collection which greatlyreduces the storage space required to process massive pointcloud data Based on this structure BloomFilter completesthe query of the first-order neighbourhood points for eachpoint and then we can calculate the area of polygon enclosedby them which greatly compresses the memory space andshortens the calculation time
323 Results of Concrete Spalling Detection The rough areasextracted on tunnel surface based on the method of rough-ness descriptor mainly contain three types of objects namelyconcrete spalling patches bolt holes and the seams betweensegments In order to identify the points belonging to spallingpatches we used the method of rough area classificationdescribed in Section 24 to separate bolt holes and the seg-ment seams from rough points so that the concrete spallingpatches can be remained
Firstly if the tunnel surface is unfolded into a planethe seam appears as a line Therefore the method of seamidentification is to project the rough point cloud extractedonto a plane and rasterize it into an image Then the Houghtransformation algorithm can be used to identify seamsbetween segments the result of which is shown in Figure 14As we can see from the figure seams including the transverseseams longitudinal seams and oblique seams have beenidentified successfully
Then the bolt holes are detected using the similarityanalysis method the separation result of which is shown inFigure 14 where the bolt holes are shown in blue blocksThe remaining patches belong to the concrete spalling areashown in red After detecting the tunnel section of 250m itwas found through statistics that the spalling patches mainlyoccurred in the subinterval between 175 and 200m so that inorder to express the detected spalling more clearly Figure 14shows the information of tunnel between 175 and 200mFinally we found seven concrete spalling patches in this case
10 Journal of Sensors
Table 2 Basic information of concrete spalling patches
No Mileage [m] Spalling patches Images Spalling area [m2] Spalling depth [m]
J1 176 00191 0112
J2 176 00276 0126
J3 177 00377 0195
J4 178 00340 0099
J5 177 00384 0129
J6 177 00681 0083
J7 193 00249 0091
study the basic information of which is shown in Table 2and the mileage position corresponding to the spalling is alsogiven at the same time
According to formula (9) the theoretical depth value andarea value of minimum spalling patch are 2mm and 10mm2respectively and the extracted results are indeed greater thanthe theoretical minimum
In this paper in order to examine the accuracy of thedetected results we conducted a jointmanual inspectionwith
the maintenance company on the seven detected spallingpatches that is each of the spalling was confirmed one by onein the tunnel It was found that each spalling did occur at thecorresponding position Therefore the false detection rate iszero and it is confirmed that the proposed concrete spallingdetection algorithm performs well In addition taking intoaccount the small size of the spalling and the large spacingbetween the points it does cause omission errors which havebeen analysed in Section 25 However these regions that
Journal of Sensors 11
Bolt holeConcrete spalling
Segment seam
7
6 5 4
32
1
Figure 14 Recognition result of bolt holes segment seams andconcrete spalling patches in tunnel subinterval of 175-200 m
could not be detected are very small and belong to the normalcategory which will not affect the performance and reliabilityof a tunnel
4 Conclusion
In this paper a new method used for concrete spallingdetection in metro tunnel from point cloud based on theroughness descriptor is proposed Firstly the point cloudacquired by mobile laser scanning system needs to eliminateoutlier points originating from ancillary facilities attachedto shield tunnel wall based on the residual error filteringalgorithmThen a roughness descriptor for the metro tunnelsurface is designed to extract the rough areas on the tunnelsurface including bolt holes segment seams and spallingpatches Finally rough area classification is performed onthe identified rough areas to accurately separate the segmentseams and bolt holes from rough areas so that the concretespalling patches are left A section of metro tunnel intervalabout 250m in Shanghai is selected to verify the validityof the proposed method and seven concrete spalling areasare detected which are identified as surface defects in metrotunnel This could be helpful for tunnel maintenance andoperation safety Compared with previous studies the con-cept of roughness descriptor is proposed to detect concretespalling which is suitable for not only flat concrete surfacesbut also nonplanar concrete surfaces and at the same timeoffer the guidance for optimal scanning parameter selection
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This research is supported by the National Science Foun-dation of China (no 41671451) the National Science andTechnologyMajor Program (2016YFB0502104) and the Fun-damental Research Funds for the Central Universities ofChina The authors would like to express appreciation tocolleagues in our laboratory for their valuable commentshelp
References
[1] T Asakura and Y Kojima ldquoTunnel maintenance in JapanrdquoTunnelling and Underground Space Technology vol 18 no 2-3pp 161ndash169 2003
[2] Y Yuan Y Bai and J Liu ldquoAssessment service state of tunnelstructurerdquo Tunnelling and Underground Space Technology vol27 no 1 pp 72ndash85 2012
[3] F Sandrone and V Labiouse ldquoIdentification and analysisof Swiss National Road tunnels pathologiesrdquo Tunnelling andUnderground Space Technology vol 26 no 2 pp 374ndash390 2011
[4] Portland Cement Association (PCA) Concrete slab surfacedefects Causes Prevention And Repair Portland CementSkokie IL USA 2001
[5] N Delatte S Chen N Maini et al ldquoApplication of non-destructive evaluation to subway tunnel systemsrdquo Transporta-tion Research Record vol 1845 no 3 pp 127ndash135 2003
[6] H Russell and J Gilmore ldquoInspection policy and proceduresfor rail transit tunnels and underground structuresrdquo TransitCooperative Research Program Synthesis of Transit Practice 1997
[7] Y Yuan X Jiang and Q Ai ldquoProbabilistic assessment forconcrete spalling in tunnel structuresrdquo ASCE-ASME Journalof Risk and Uncertainty in Engineering Systems Part A CivilEngineering vol 3 no 4 2017
[8] M-K Kim H Sohn and C-C Chang ldquoLocalization andquantification of concrete spalling defects using terrestrial laserscanningrdquo Journal of Computing in Civil Engineering vol 29 no6 2015
[9] Z Zhu and I Bilakis ldquoDetecting air pockets for architecturalconcrete quality assessment using visual sensingrdquo ElectronicJournal of Information Technology in Construction vol 13 pp86ndash102 2008
[10] American Concrete Institute ACI Manual of Concrete Inspec-tion ACI Committee 311 SP-2(07) Detroit Mich USA 2007
[11] A M Paterson G R Dowling and D A ChamberlainldquoBuilding inspection can computer vision helprdquo Automationin Construction vol 7 no 1 pp 13ndash20 1997
[12] B Guldur Erkal and J F Hajjar ldquoLaser-based surface damagedetection and quantification using predicted surface proper-tiesrdquo Automation in Construction vol 83 pp 285ndash302 2017
[13] T Dawood Z Zhu and T Zayed ldquoMachine vision-basedmodelfor spalling detection and quantification in subway networksrdquoAutomation in Construction vol 81 pp 149ndash160 2017
[14] RMedina J Llamas J Gomez-Garcıa-Bermejo E Zalama andM Segarra ldquoCrack detection in concrete tunnels using a Gaborfilter invariant to rotationrdquo Sensors vol 17 no 7 p 1670 2017
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Journal of Sensors 7
Figure 8A scan line of tunnel point cloudTheblack ring representsthe real position of a section on the tunnel inner wall and thedeviation of red points away from black ring is regarded as theprecision of point cloud (Δ)
value of roughness descriptor is 1 When h gt 0 the value ofroughness descriptor is greater than 1 It indicates that theposition is rough relative to the normal position and may bespalling
Therefore the detectability of concrete spalling basedon the roughness descriptor is determined by the spallingdepth ℎ and the point spacingm while these two parametersare mainly affected by the instrument accuracy and theset parameters of the MLS system used when collectingpoint cloud in metro tunnel namely range error Δ and theresolution of laser scanner as well as the running velocityof MLS system Firstly the range error Δ of laser scannerindicates the precision of the collected point cloud of metrotunnel Taking out a scan line of point cloud and expanding itinto a straight line as shown by red points in Figure 8 assumethat the black ring is the real position of a section on thetunnel inner wall while the deviation of captured points awayfrom the innerwall is regarded as the precision of point cloudwhich is represented byΔThus it can be seen that the spallingpatches will not be detected when the value of spalling depthℎ is less than Δ
The other factor that affects the detectability of spallingis the point spacing m including the vertical spacing andlongitudinal spacing The vertical spacing of point clouddepends on the resolution of the scanner When settingdifferent resolutions the number of scanning points on theone scan-line changes accordingly In addition since thefrequency of scanner is usually fixed the velocity of themobile laser scanning system determines the point spacingin the direction of the mileage commonly referred to asthe longitudinal spacing Taking a microelement on thesurface of tunnel as an example in Figure 9 the verticaland longitudinal spacing of point cloud are represented by1198981 and 1198982 respectively and the blue areas are used toindicate the spalling patches It follows that when the areaof spalling patches less than the product of vertical spacingand longitudinal spacing it cannot be detected either
Therefore when the depth ℎ and the area 119878119888 of a concretespalling patch satisfy the following formula (9) it can beextracted from the point cloud of tunnel surface whichcan also be used as a guideline to select optimal scanningparameters for MLS system
ℎ gt 119878119888 gt 1198981 lowast 1198982 (9)
m1
m2
Figure 9 A microelement on tunnel surface The red dots indicatethe points on a microelement of tunnel surface captured by scannerand the blue areas indicate the spalling patches
Figure 10 Mobile laser scanning system
3 Case Study
31 Data Collection of Metro Tunnel A section of a metrotunnel in Shanghai was selected as the experimental area witha total length of about 250 meters The mobile laser scanning(MLS) system is equippedwith a scanner of FAROFOCUS3DX330 for point cloud data collection in the tunnel as shownin Figure 10 the scanner of which has a scanning range of300∘ and working frequency of 100Hz In order not to affectthe routine operation of the subway the experimental dataacquisition was carried out between midnight and three inthe morning And the resolution of scanner is set to 14 sothe number of points in one circular scan line is about 9760and the vertical point spacing 1198981 is about 2mm Generallyduring the period of data acquisition to ensure the densityof point cloud the running velocity of MLS system on thesubway track is set to 05ms so the average point spacing1198982 of the collected point clouds in the mileage direction isabout 5mm The general information of the case area anddata collection is shown in Table 1
32 Experimental Results
321 Outlier Removal Result of Tunnel Point Cloud Thecaptured point cloud data by MLS system mainly containsthe information of the tunnel surface where it is also mixedwith some outlier points originating from the subway trackscables lighting equipment and other facilities which willinevitably have great interference on the concrete spalling
8 Journal of Sensors
Resid
ual e
rror
Point serial number
004
002
0
minus002
minus004
minus006
minus008
minus01
minus0120 500 1000 1500 2000 2500 3000 3500
(a)
(b) (c)
Figure 11 Outlier removal (a) Residual error curve of one scan line (b) before the outlier removal and (c) after the outlier removal
Table 1 General information of the case area and data collection
Parameters ValueLength of case tunnel 250 mRadius of the tunnel 275 mAverage velocity during data collection 05 msScanning distance 330 mScanning range 300∘
Working frequency 100 HzResolution 14Range error () 2 mmVertical point spacing (1198981) 2 mmLongitudinal point spacing (1198982) 5 mmPoint density 100000 ptsm2
Total points gt 500000000 pts
identification Thus according to the outlier points removalalgorithm introduced in Section 22 the residual error curvefor each circular scan line can be generated an example ofwhich is shown in Figure 11(a) Since the fluctuation range ofresidual errors is between plus and minus 001 hence it canbe seen that points with an error of less than negative 001 canbe considered outliers and then eliminated Figures 11(b) and11(c) show the point cloud of tunnel before and after outlierpoints removal respectively
Most of the outlier points with a certain distance fromthe tunnel surface can be removed using the residual errorfiltering algorithm However there are still a small fractionof points from the bottom of pipeline facility that clings tothe tunnel inner wall and cannot be completely eliminatedthrough thismethod causing them to eventually be identifiedas rough areas This part of points is usually presented in theform of a line so it can be identified and further removedtogether with ring seams through the algorithm of Houghtransformation later
322 Rough Area Extraction After removing the outlierpoints from original point cloud data of metro tunnel aroughness descriptor based method is applied to extract thepoints of rough areas on tunnel surface for the purpose offurther identifying the concrete spalling patches therefromThus in order to calculate the value of roughness descriptorfor each point triangular meshes are first constructed forboth the remaining points after outlier removal and thecorresponding projected points on the cylindrical surfacethereby obtaining the surface area and projected area ofthe first-order neighbourhood around each point Thenaccording to the definition of roughness descriptor the ratioof surface area to the projected area around each point iscalculated a histogram of which is also generated indicatingthe number of points corresponding to different roughnessvalues as shown in Figure 12
Journal of Sensors 9
times105
poin
t num
ber
roughness descriptor
2
18
16
14
12
1
08
06
04
02
009 1 11 12 13 14 15 16 17
Figure 12 Histogram of roughness descriptor
Figure 13 Roughness map of tunnel surface
According to the value of the roughness descriptor foreach point obtained we need to determine an accuratethreshold to extract the points belonging to rough areasTaking into account the scanner accuracy and the set param-eters of MLS system during operation in this experimentmainly relying on the range error and the point spacing119898 the expression of roughness descriptor can be furtherrepresented as follows
It can be seen from formula (10) that when the value ofthe depth ℎ for a certain point tends to zero roughness ratiois at a critical condition Therefore with the value of pointspacing119898 and range error Δ in this case study the thresholdof roughness descriptor has been obtained as about 105 andpoints with proportion greater than 105 are filtered as roughareas Expand the rough point cloud on the tunnel surfaceinto a plane as shown in Figure 13
It should be noted that we cannot thin the original pointcloud otherwise some points belonging to rough areas maybe missed which will affect the detection of spalling damageHowever when calculating the roughness value of each pointthe number of tunnel point clouds captured byMLS system isvery huge which will take a lot of time to find a polygon area
surrounded by its first-order neighbourhood points for eachpoint Accordingly BitMap and BloomFilter are adopted inthis part to improve performing efficiency of the algorithmBitmap is a compact data storage structure that allocates 1 bitof memory for each element in the collection which greatlyreduces the storage space required to process massive pointcloud data Based on this structure BloomFilter completesthe query of the first-order neighbourhood points for eachpoint and then we can calculate the area of polygon enclosedby them which greatly compresses the memory space andshortens the calculation time
323 Results of Concrete Spalling Detection The rough areasextracted on tunnel surface based on the method of rough-ness descriptor mainly contain three types of objects namelyconcrete spalling patches bolt holes and the seams betweensegments In order to identify the points belonging to spallingpatches we used the method of rough area classificationdescribed in Section 24 to separate bolt holes and the seg-ment seams from rough points so that the concrete spallingpatches can be remained
Firstly if the tunnel surface is unfolded into a planethe seam appears as a line Therefore the method of seamidentification is to project the rough point cloud extractedonto a plane and rasterize it into an image Then the Houghtransformation algorithm can be used to identify seamsbetween segments the result of which is shown in Figure 14As we can see from the figure seams including the transverseseams longitudinal seams and oblique seams have beenidentified successfully
Then the bolt holes are detected using the similarityanalysis method the separation result of which is shown inFigure 14 where the bolt holes are shown in blue blocksThe remaining patches belong to the concrete spalling areashown in red After detecting the tunnel section of 250m itwas found through statistics that the spalling patches mainlyoccurred in the subinterval between 175 and 200m so that inorder to express the detected spalling more clearly Figure 14shows the information of tunnel between 175 and 200mFinally we found seven concrete spalling patches in this case
10 Journal of Sensors
Table 2 Basic information of concrete spalling patches
No Mileage [m] Spalling patches Images Spalling area [m2] Spalling depth [m]
J1 176 00191 0112
J2 176 00276 0126
J3 177 00377 0195
J4 178 00340 0099
J5 177 00384 0129
J6 177 00681 0083
J7 193 00249 0091
study the basic information of which is shown in Table 2and the mileage position corresponding to the spalling is alsogiven at the same time
According to formula (9) the theoretical depth value andarea value of minimum spalling patch are 2mm and 10mm2respectively and the extracted results are indeed greater thanthe theoretical minimum
In this paper in order to examine the accuracy of thedetected results we conducted a jointmanual inspectionwith
the maintenance company on the seven detected spallingpatches that is each of the spalling was confirmed one by onein the tunnel It was found that each spalling did occur at thecorresponding position Therefore the false detection rate iszero and it is confirmed that the proposed concrete spallingdetection algorithm performs well In addition taking intoaccount the small size of the spalling and the large spacingbetween the points it does cause omission errors which havebeen analysed in Section 25 However these regions that
Journal of Sensors 11
Bolt holeConcrete spalling
Segment seam
7
6 5 4
32
1
Figure 14 Recognition result of bolt holes segment seams andconcrete spalling patches in tunnel subinterval of 175-200 m
could not be detected are very small and belong to the normalcategory which will not affect the performance and reliabilityof a tunnel
4 Conclusion
In this paper a new method used for concrete spallingdetection in metro tunnel from point cloud based on theroughness descriptor is proposed Firstly the point cloudacquired by mobile laser scanning system needs to eliminateoutlier points originating from ancillary facilities attachedto shield tunnel wall based on the residual error filteringalgorithmThen a roughness descriptor for the metro tunnelsurface is designed to extract the rough areas on the tunnelsurface including bolt holes segment seams and spallingpatches Finally rough area classification is performed onthe identified rough areas to accurately separate the segmentseams and bolt holes from rough areas so that the concretespalling patches are left A section of metro tunnel intervalabout 250m in Shanghai is selected to verify the validityof the proposed method and seven concrete spalling areasare detected which are identified as surface defects in metrotunnel This could be helpful for tunnel maintenance andoperation safety Compared with previous studies the con-cept of roughness descriptor is proposed to detect concretespalling which is suitable for not only flat concrete surfacesbut also nonplanar concrete surfaces and at the same timeoffer the guidance for optimal scanning parameter selection
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This research is supported by the National Science Foun-dation of China (no 41671451) the National Science andTechnologyMajor Program (2016YFB0502104) and the Fun-damental Research Funds for the Central Universities ofChina The authors would like to express appreciation tocolleagues in our laboratory for their valuable commentshelp
References
[1] T Asakura and Y Kojima ldquoTunnel maintenance in JapanrdquoTunnelling and Underground Space Technology vol 18 no 2-3pp 161ndash169 2003
[2] Y Yuan Y Bai and J Liu ldquoAssessment service state of tunnelstructurerdquo Tunnelling and Underground Space Technology vol27 no 1 pp 72ndash85 2012
[3] F Sandrone and V Labiouse ldquoIdentification and analysisof Swiss National Road tunnels pathologiesrdquo Tunnelling andUnderground Space Technology vol 26 no 2 pp 374ndash390 2011
[4] Portland Cement Association (PCA) Concrete slab surfacedefects Causes Prevention And Repair Portland CementSkokie IL USA 2001
[5] N Delatte S Chen N Maini et al ldquoApplication of non-destructive evaluation to subway tunnel systemsrdquo Transporta-tion Research Record vol 1845 no 3 pp 127ndash135 2003
[6] H Russell and J Gilmore ldquoInspection policy and proceduresfor rail transit tunnels and underground structuresrdquo TransitCooperative Research Program Synthesis of Transit Practice 1997
[7] Y Yuan X Jiang and Q Ai ldquoProbabilistic assessment forconcrete spalling in tunnel structuresrdquo ASCE-ASME Journalof Risk and Uncertainty in Engineering Systems Part A CivilEngineering vol 3 no 4 2017
[8] M-K Kim H Sohn and C-C Chang ldquoLocalization andquantification of concrete spalling defects using terrestrial laserscanningrdquo Journal of Computing in Civil Engineering vol 29 no6 2015
[9] Z Zhu and I Bilakis ldquoDetecting air pockets for architecturalconcrete quality assessment using visual sensingrdquo ElectronicJournal of Information Technology in Construction vol 13 pp86ndash102 2008
[10] American Concrete Institute ACI Manual of Concrete Inspec-tion ACI Committee 311 SP-2(07) Detroit Mich USA 2007
[11] A M Paterson G R Dowling and D A ChamberlainldquoBuilding inspection can computer vision helprdquo Automationin Construction vol 7 no 1 pp 13ndash20 1997
[12] B Guldur Erkal and J F Hajjar ldquoLaser-based surface damagedetection and quantification using predicted surface proper-tiesrdquo Automation in Construction vol 83 pp 285ndash302 2017
[13] T Dawood Z Zhu and T Zayed ldquoMachine vision-basedmodelfor spalling detection and quantification in subway networksrdquoAutomation in Construction vol 81 pp 149ndash160 2017
[14] RMedina J Llamas J Gomez-Garcıa-Bermejo E Zalama andM Segarra ldquoCrack detection in concrete tunnels using a Gaborfilter invariant to rotationrdquo Sensors vol 17 no 7 p 1670 2017
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
8 Journal of Sensors
Resid
ual e
rror
Point serial number
004
002
0
minus002
minus004
minus006
minus008
minus01
minus0120 500 1000 1500 2000 2500 3000 3500
(a)
(b) (c)
Figure 11 Outlier removal (a) Residual error curve of one scan line (b) before the outlier removal and (c) after the outlier removal
Table 1 General information of the case area and data collection
Parameters ValueLength of case tunnel 250 mRadius of the tunnel 275 mAverage velocity during data collection 05 msScanning distance 330 mScanning range 300∘
Working frequency 100 HzResolution 14Range error () 2 mmVertical point spacing (1198981) 2 mmLongitudinal point spacing (1198982) 5 mmPoint density 100000 ptsm2
Total points gt 500000000 pts
identification Thus according to the outlier points removalalgorithm introduced in Section 22 the residual error curvefor each circular scan line can be generated an example ofwhich is shown in Figure 11(a) Since the fluctuation range ofresidual errors is between plus and minus 001 hence it canbe seen that points with an error of less than negative 001 canbe considered outliers and then eliminated Figures 11(b) and11(c) show the point cloud of tunnel before and after outlierpoints removal respectively
Most of the outlier points with a certain distance fromthe tunnel surface can be removed using the residual errorfiltering algorithm However there are still a small fractionof points from the bottom of pipeline facility that clings tothe tunnel inner wall and cannot be completely eliminatedthrough thismethod causing them to eventually be identifiedas rough areas This part of points is usually presented in theform of a line so it can be identified and further removedtogether with ring seams through the algorithm of Houghtransformation later
322 Rough Area Extraction After removing the outlierpoints from original point cloud data of metro tunnel aroughness descriptor based method is applied to extract thepoints of rough areas on tunnel surface for the purpose offurther identifying the concrete spalling patches therefromThus in order to calculate the value of roughness descriptorfor each point triangular meshes are first constructed forboth the remaining points after outlier removal and thecorresponding projected points on the cylindrical surfacethereby obtaining the surface area and projected area ofthe first-order neighbourhood around each point Thenaccording to the definition of roughness descriptor the ratioof surface area to the projected area around each point iscalculated a histogram of which is also generated indicatingthe number of points corresponding to different roughnessvalues as shown in Figure 12
Journal of Sensors 9
times105
poin
t num
ber
roughness descriptor
2
18
16
14
12
1
08
06
04
02
009 1 11 12 13 14 15 16 17
Figure 12 Histogram of roughness descriptor
Figure 13 Roughness map of tunnel surface
According to the value of the roughness descriptor foreach point obtained we need to determine an accuratethreshold to extract the points belonging to rough areasTaking into account the scanner accuracy and the set param-eters of MLS system during operation in this experimentmainly relying on the range error and the point spacing119898 the expression of roughness descriptor can be furtherrepresented as follows
It can be seen from formula (10) that when the value ofthe depth ℎ for a certain point tends to zero roughness ratiois at a critical condition Therefore with the value of pointspacing119898 and range error Δ in this case study the thresholdof roughness descriptor has been obtained as about 105 andpoints with proportion greater than 105 are filtered as roughareas Expand the rough point cloud on the tunnel surfaceinto a plane as shown in Figure 13
It should be noted that we cannot thin the original pointcloud otherwise some points belonging to rough areas maybe missed which will affect the detection of spalling damageHowever when calculating the roughness value of each pointthe number of tunnel point clouds captured byMLS system isvery huge which will take a lot of time to find a polygon area
surrounded by its first-order neighbourhood points for eachpoint Accordingly BitMap and BloomFilter are adopted inthis part to improve performing efficiency of the algorithmBitmap is a compact data storage structure that allocates 1 bitof memory for each element in the collection which greatlyreduces the storage space required to process massive pointcloud data Based on this structure BloomFilter completesthe query of the first-order neighbourhood points for eachpoint and then we can calculate the area of polygon enclosedby them which greatly compresses the memory space andshortens the calculation time
323 Results of Concrete Spalling Detection The rough areasextracted on tunnel surface based on the method of rough-ness descriptor mainly contain three types of objects namelyconcrete spalling patches bolt holes and the seams betweensegments In order to identify the points belonging to spallingpatches we used the method of rough area classificationdescribed in Section 24 to separate bolt holes and the seg-ment seams from rough points so that the concrete spallingpatches can be remained
Firstly if the tunnel surface is unfolded into a planethe seam appears as a line Therefore the method of seamidentification is to project the rough point cloud extractedonto a plane and rasterize it into an image Then the Houghtransformation algorithm can be used to identify seamsbetween segments the result of which is shown in Figure 14As we can see from the figure seams including the transverseseams longitudinal seams and oblique seams have beenidentified successfully
Then the bolt holes are detected using the similarityanalysis method the separation result of which is shown inFigure 14 where the bolt holes are shown in blue blocksThe remaining patches belong to the concrete spalling areashown in red After detecting the tunnel section of 250m itwas found through statistics that the spalling patches mainlyoccurred in the subinterval between 175 and 200m so that inorder to express the detected spalling more clearly Figure 14shows the information of tunnel between 175 and 200mFinally we found seven concrete spalling patches in this case
10 Journal of Sensors
Table 2 Basic information of concrete spalling patches
No Mileage [m] Spalling patches Images Spalling area [m2] Spalling depth [m]
J1 176 00191 0112
J2 176 00276 0126
J3 177 00377 0195
J4 178 00340 0099
J5 177 00384 0129
J6 177 00681 0083
J7 193 00249 0091
study the basic information of which is shown in Table 2and the mileage position corresponding to the spalling is alsogiven at the same time
According to formula (9) the theoretical depth value andarea value of minimum spalling patch are 2mm and 10mm2respectively and the extracted results are indeed greater thanthe theoretical minimum
In this paper in order to examine the accuracy of thedetected results we conducted a jointmanual inspectionwith
the maintenance company on the seven detected spallingpatches that is each of the spalling was confirmed one by onein the tunnel It was found that each spalling did occur at thecorresponding position Therefore the false detection rate iszero and it is confirmed that the proposed concrete spallingdetection algorithm performs well In addition taking intoaccount the small size of the spalling and the large spacingbetween the points it does cause omission errors which havebeen analysed in Section 25 However these regions that
Journal of Sensors 11
Bolt holeConcrete spalling
Segment seam
7
6 5 4
32
1
Figure 14 Recognition result of bolt holes segment seams andconcrete spalling patches in tunnel subinterval of 175-200 m
could not be detected are very small and belong to the normalcategory which will not affect the performance and reliabilityof a tunnel
4 Conclusion
In this paper a new method used for concrete spallingdetection in metro tunnel from point cloud based on theroughness descriptor is proposed Firstly the point cloudacquired by mobile laser scanning system needs to eliminateoutlier points originating from ancillary facilities attachedto shield tunnel wall based on the residual error filteringalgorithmThen a roughness descriptor for the metro tunnelsurface is designed to extract the rough areas on the tunnelsurface including bolt holes segment seams and spallingpatches Finally rough area classification is performed onthe identified rough areas to accurately separate the segmentseams and bolt holes from rough areas so that the concretespalling patches are left A section of metro tunnel intervalabout 250m in Shanghai is selected to verify the validityof the proposed method and seven concrete spalling areasare detected which are identified as surface defects in metrotunnel This could be helpful for tunnel maintenance andoperation safety Compared with previous studies the con-cept of roughness descriptor is proposed to detect concretespalling which is suitable for not only flat concrete surfacesbut also nonplanar concrete surfaces and at the same timeoffer the guidance for optimal scanning parameter selection
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This research is supported by the National Science Foun-dation of China (no 41671451) the National Science andTechnologyMajor Program (2016YFB0502104) and the Fun-damental Research Funds for the Central Universities ofChina The authors would like to express appreciation tocolleagues in our laboratory for their valuable commentshelp
References
[1] T Asakura and Y Kojima ldquoTunnel maintenance in JapanrdquoTunnelling and Underground Space Technology vol 18 no 2-3pp 161ndash169 2003
[2] Y Yuan Y Bai and J Liu ldquoAssessment service state of tunnelstructurerdquo Tunnelling and Underground Space Technology vol27 no 1 pp 72ndash85 2012
[3] F Sandrone and V Labiouse ldquoIdentification and analysisof Swiss National Road tunnels pathologiesrdquo Tunnelling andUnderground Space Technology vol 26 no 2 pp 374ndash390 2011
[4] Portland Cement Association (PCA) Concrete slab surfacedefects Causes Prevention And Repair Portland CementSkokie IL USA 2001
[5] N Delatte S Chen N Maini et al ldquoApplication of non-destructive evaluation to subway tunnel systemsrdquo Transporta-tion Research Record vol 1845 no 3 pp 127ndash135 2003
[6] H Russell and J Gilmore ldquoInspection policy and proceduresfor rail transit tunnels and underground structuresrdquo TransitCooperative Research Program Synthesis of Transit Practice 1997
[7] Y Yuan X Jiang and Q Ai ldquoProbabilistic assessment forconcrete spalling in tunnel structuresrdquo ASCE-ASME Journalof Risk and Uncertainty in Engineering Systems Part A CivilEngineering vol 3 no 4 2017
[8] M-K Kim H Sohn and C-C Chang ldquoLocalization andquantification of concrete spalling defects using terrestrial laserscanningrdquo Journal of Computing in Civil Engineering vol 29 no6 2015
[9] Z Zhu and I Bilakis ldquoDetecting air pockets for architecturalconcrete quality assessment using visual sensingrdquo ElectronicJournal of Information Technology in Construction vol 13 pp86ndash102 2008
[10] American Concrete Institute ACI Manual of Concrete Inspec-tion ACI Committee 311 SP-2(07) Detroit Mich USA 2007
[11] A M Paterson G R Dowling and D A ChamberlainldquoBuilding inspection can computer vision helprdquo Automationin Construction vol 7 no 1 pp 13ndash20 1997
[12] B Guldur Erkal and J F Hajjar ldquoLaser-based surface damagedetection and quantification using predicted surface proper-tiesrdquo Automation in Construction vol 83 pp 285ndash302 2017
[13] T Dawood Z Zhu and T Zayed ldquoMachine vision-basedmodelfor spalling detection and quantification in subway networksrdquoAutomation in Construction vol 81 pp 149ndash160 2017
[14] RMedina J Llamas J Gomez-Garcıa-Bermejo E Zalama andM Segarra ldquoCrack detection in concrete tunnels using a Gaborfilter invariant to rotationrdquo Sensors vol 17 no 7 p 1670 2017
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Journal of Sensors 9
times105
poin
t num
ber
roughness descriptor
2
18
16
14
12
1
08
06
04
02
009 1 11 12 13 14 15 16 17
Figure 12 Histogram of roughness descriptor
Figure 13 Roughness map of tunnel surface
According to the value of the roughness descriptor foreach point obtained we need to determine an accuratethreshold to extract the points belonging to rough areasTaking into account the scanner accuracy and the set param-eters of MLS system during operation in this experimentmainly relying on the range error and the point spacing119898 the expression of roughness descriptor can be furtherrepresented as follows
It can be seen from formula (10) that when the value ofthe depth ℎ for a certain point tends to zero roughness ratiois at a critical condition Therefore with the value of pointspacing119898 and range error Δ in this case study the thresholdof roughness descriptor has been obtained as about 105 andpoints with proportion greater than 105 are filtered as roughareas Expand the rough point cloud on the tunnel surfaceinto a plane as shown in Figure 13
It should be noted that we cannot thin the original pointcloud otherwise some points belonging to rough areas maybe missed which will affect the detection of spalling damageHowever when calculating the roughness value of each pointthe number of tunnel point clouds captured byMLS system isvery huge which will take a lot of time to find a polygon area
surrounded by its first-order neighbourhood points for eachpoint Accordingly BitMap and BloomFilter are adopted inthis part to improve performing efficiency of the algorithmBitmap is a compact data storage structure that allocates 1 bitof memory for each element in the collection which greatlyreduces the storage space required to process massive pointcloud data Based on this structure BloomFilter completesthe query of the first-order neighbourhood points for eachpoint and then we can calculate the area of polygon enclosedby them which greatly compresses the memory space andshortens the calculation time
323 Results of Concrete Spalling Detection The rough areasextracted on tunnel surface based on the method of rough-ness descriptor mainly contain three types of objects namelyconcrete spalling patches bolt holes and the seams betweensegments In order to identify the points belonging to spallingpatches we used the method of rough area classificationdescribed in Section 24 to separate bolt holes and the seg-ment seams from rough points so that the concrete spallingpatches can be remained
Firstly if the tunnel surface is unfolded into a planethe seam appears as a line Therefore the method of seamidentification is to project the rough point cloud extractedonto a plane and rasterize it into an image Then the Houghtransformation algorithm can be used to identify seamsbetween segments the result of which is shown in Figure 14As we can see from the figure seams including the transverseseams longitudinal seams and oblique seams have beenidentified successfully
Then the bolt holes are detected using the similarityanalysis method the separation result of which is shown inFigure 14 where the bolt holes are shown in blue blocksThe remaining patches belong to the concrete spalling areashown in red After detecting the tunnel section of 250m itwas found through statistics that the spalling patches mainlyoccurred in the subinterval between 175 and 200m so that inorder to express the detected spalling more clearly Figure 14shows the information of tunnel between 175 and 200mFinally we found seven concrete spalling patches in this case
10 Journal of Sensors
Table 2 Basic information of concrete spalling patches
No Mileage [m] Spalling patches Images Spalling area [m2] Spalling depth [m]
J1 176 00191 0112
J2 176 00276 0126
J3 177 00377 0195
J4 178 00340 0099
J5 177 00384 0129
J6 177 00681 0083
J7 193 00249 0091
study the basic information of which is shown in Table 2and the mileage position corresponding to the spalling is alsogiven at the same time
According to formula (9) the theoretical depth value andarea value of minimum spalling patch are 2mm and 10mm2respectively and the extracted results are indeed greater thanthe theoretical minimum
In this paper in order to examine the accuracy of thedetected results we conducted a jointmanual inspectionwith
the maintenance company on the seven detected spallingpatches that is each of the spalling was confirmed one by onein the tunnel It was found that each spalling did occur at thecorresponding position Therefore the false detection rate iszero and it is confirmed that the proposed concrete spallingdetection algorithm performs well In addition taking intoaccount the small size of the spalling and the large spacingbetween the points it does cause omission errors which havebeen analysed in Section 25 However these regions that
Journal of Sensors 11
Bolt holeConcrete spalling
Segment seam
7
6 5 4
32
1
Figure 14 Recognition result of bolt holes segment seams andconcrete spalling patches in tunnel subinterval of 175-200 m
could not be detected are very small and belong to the normalcategory which will not affect the performance and reliabilityof a tunnel
4 Conclusion
In this paper a new method used for concrete spallingdetection in metro tunnel from point cloud based on theroughness descriptor is proposed Firstly the point cloudacquired by mobile laser scanning system needs to eliminateoutlier points originating from ancillary facilities attachedto shield tunnel wall based on the residual error filteringalgorithmThen a roughness descriptor for the metro tunnelsurface is designed to extract the rough areas on the tunnelsurface including bolt holes segment seams and spallingpatches Finally rough area classification is performed onthe identified rough areas to accurately separate the segmentseams and bolt holes from rough areas so that the concretespalling patches are left A section of metro tunnel intervalabout 250m in Shanghai is selected to verify the validityof the proposed method and seven concrete spalling areasare detected which are identified as surface defects in metrotunnel This could be helpful for tunnel maintenance andoperation safety Compared with previous studies the con-cept of roughness descriptor is proposed to detect concretespalling which is suitable for not only flat concrete surfacesbut also nonplanar concrete surfaces and at the same timeoffer the guidance for optimal scanning parameter selection
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This research is supported by the National Science Foun-dation of China (no 41671451) the National Science andTechnologyMajor Program (2016YFB0502104) and the Fun-damental Research Funds for the Central Universities ofChina The authors would like to express appreciation tocolleagues in our laboratory for their valuable commentshelp
References
[1] T Asakura and Y Kojima ldquoTunnel maintenance in JapanrdquoTunnelling and Underground Space Technology vol 18 no 2-3pp 161ndash169 2003
[2] Y Yuan Y Bai and J Liu ldquoAssessment service state of tunnelstructurerdquo Tunnelling and Underground Space Technology vol27 no 1 pp 72ndash85 2012
[3] F Sandrone and V Labiouse ldquoIdentification and analysisof Swiss National Road tunnels pathologiesrdquo Tunnelling andUnderground Space Technology vol 26 no 2 pp 374ndash390 2011
[4] Portland Cement Association (PCA) Concrete slab surfacedefects Causes Prevention And Repair Portland CementSkokie IL USA 2001
[5] N Delatte S Chen N Maini et al ldquoApplication of non-destructive evaluation to subway tunnel systemsrdquo Transporta-tion Research Record vol 1845 no 3 pp 127ndash135 2003
[6] H Russell and J Gilmore ldquoInspection policy and proceduresfor rail transit tunnels and underground structuresrdquo TransitCooperative Research Program Synthesis of Transit Practice 1997
[7] Y Yuan X Jiang and Q Ai ldquoProbabilistic assessment forconcrete spalling in tunnel structuresrdquo ASCE-ASME Journalof Risk and Uncertainty in Engineering Systems Part A CivilEngineering vol 3 no 4 2017
[8] M-K Kim H Sohn and C-C Chang ldquoLocalization andquantification of concrete spalling defects using terrestrial laserscanningrdquo Journal of Computing in Civil Engineering vol 29 no6 2015
[9] Z Zhu and I Bilakis ldquoDetecting air pockets for architecturalconcrete quality assessment using visual sensingrdquo ElectronicJournal of Information Technology in Construction vol 13 pp86ndash102 2008
[10] American Concrete Institute ACI Manual of Concrete Inspec-tion ACI Committee 311 SP-2(07) Detroit Mich USA 2007
[11] A M Paterson G R Dowling and D A ChamberlainldquoBuilding inspection can computer vision helprdquo Automationin Construction vol 7 no 1 pp 13ndash20 1997
[12] B Guldur Erkal and J F Hajjar ldquoLaser-based surface damagedetection and quantification using predicted surface proper-tiesrdquo Automation in Construction vol 83 pp 285ndash302 2017
[13] T Dawood Z Zhu and T Zayed ldquoMachine vision-basedmodelfor spalling detection and quantification in subway networksrdquoAutomation in Construction vol 81 pp 149ndash160 2017
[14] RMedina J Llamas J Gomez-Garcıa-Bermejo E Zalama andM Segarra ldquoCrack detection in concrete tunnels using a Gaborfilter invariant to rotationrdquo Sensors vol 17 no 7 p 1670 2017
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
10 Journal of Sensors
Table 2 Basic information of concrete spalling patches
No Mileage [m] Spalling patches Images Spalling area [m2] Spalling depth [m]
J1 176 00191 0112
J2 176 00276 0126
J3 177 00377 0195
J4 178 00340 0099
J5 177 00384 0129
J6 177 00681 0083
J7 193 00249 0091
study the basic information of which is shown in Table 2and the mileage position corresponding to the spalling is alsogiven at the same time
According to formula (9) the theoretical depth value andarea value of minimum spalling patch are 2mm and 10mm2respectively and the extracted results are indeed greater thanthe theoretical minimum
In this paper in order to examine the accuracy of thedetected results we conducted a jointmanual inspectionwith
the maintenance company on the seven detected spallingpatches that is each of the spalling was confirmed one by onein the tunnel It was found that each spalling did occur at thecorresponding position Therefore the false detection rate iszero and it is confirmed that the proposed concrete spallingdetection algorithm performs well In addition taking intoaccount the small size of the spalling and the large spacingbetween the points it does cause omission errors which havebeen analysed in Section 25 However these regions that
Journal of Sensors 11
Bolt holeConcrete spalling
Segment seam
7
6 5 4
32
1
Figure 14 Recognition result of bolt holes segment seams andconcrete spalling patches in tunnel subinterval of 175-200 m
could not be detected are very small and belong to the normalcategory which will not affect the performance and reliabilityof a tunnel
4 Conclusion
In this paper a new method used for concrete spallingdetection in metro tunnel from point cloud based on theroughness descriptor is proposed Firstly the point cloudacquired by mobile laser scanning system needs to eliminateoutlier points originating from ancillary facilities attachedto shield tunnel wall based on the residual error filteringalgorithmThen a roughness descriptor for the metro tunnelsurface is designed to extract the rough areas on the tunnelsurface including bolt holes segment seams and spallingpatches Finally rough area classification is performed onthe identified rough areas to accurately separate the segmentseams and bolt holes from rough areas so that the concretespalling patches are left A section of metro tunnel intervalabout 250m in Shanghai is selected to verify the validityof the proposed method and seven concrete spalling areasare detected which are identified as surface defects in metrotunnel This could be helpful for tunnel maintenance andoperation safety Compared with previous studies the con-cept of roughness descriptor is proposed to detect concretespalling which is suitable for not only flat concrete surfacesbut also nonplanar concrete surfaces and at the same timeoffer the guidance for optimal scanning parameter selection
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This research is supported by the National Science Foun-dation of China (no 41671451) the National Science andTechnologyMajor Program (2016YFB0502104) and the Fun-damental Research Funds for the Central Universities ofChina The authors would like to express appreciation tocolleagues in our laboratory for their valuable commentshelp
References
[1] T Asakura and Y Kojima ldquoTunnel maintenance in JapanrdquoTunnelling and Underground Space Technology vol 18 no 2-3pp 161ndash169 2003
[2] Y Yuan Y Bai and J Liu ldquoAssessment service state of tunnelstructurerdquo Tunnelling and Underground Space Technology vol27 no 1 pp 72ndash85 2012
[3] F Sandrone and V Labiouse ldquoIdentification and analysisof Swiss National Road tunnels pathologiesrdquo Tunnelling andUnderground Space Technology vol 26 no 2 pp 374ndash390 2011
[4] Portland Cement Association (PCA) Concrete slab surfacedefects Causes Prevention And Repair Portland CementSkokie IL USA 2001
[5] N Delatte S Chen N Maini et al ldquoApplication of non-destructive evaluation to subway tunnel systemsrdquo Transporta-tion Research Record vol 1845 no 3 pp 127ndash135 2003
[6] H Russell and J Gilmore ldquoInspection policy and proceduresfor rail transit tunnels and underground structuresrdquo TransitCooperative Research Program Synthesis of Transit Practice 1997
[7] Y Yuan X Jiang and Q Ai ldquoProbabilistic assessment forconcrete spalling in tunnel structuresrdquo ASCE-ASME Journalof Risk and Uncertainty in Engineering Systems Part A CivilEngineering vol 3 no 4 2017
[8] M-K Kim H Sohn and C-C Chang ldquoLocalization andquantification of concrete spalling defects using terrestrial laserscanningrdquo Journal of Computing in Civil Engineering vol 29 no6 2015
[9] Z Zhu and I Bilakis ldquoDetecting air pockets for architecturalconcrete quality assessment using visual sensingrdquo ElectronicJournal of Information Technology in Construction vol 13 pp86ndash102 2008
[10] American Concrete Institute ACI Manual of Concrete Inspec-tion ACI Committee 311 SP-2(07) Detroit Mich USA 2007
[11] A M Paterson G R Dowling and D A ChamberlainldquoBuilding inspection can computer vision helprdquo Automationin Construction vol 7 no 1 pp 13ndash20 1997
[12] B Guldur Erkal and J F Hajjar ldquoLaser-based surface damagedetection and quantification using predicted surface proper-tiesrdquo Automation in Construction vol 83 pp 285ndash302 2017
[13] T Dawood Z Zhu and T Zayed ldquoMachine vision-basedmodelfor spalling detection and quantification in subway networksrdquoAutomation in Construction vol 81 pp 149ndash160 2017
[14] RMedina J Llamas J Gomez-Garcıa-Bermejo E Zalama andM Segarra ldquoCrack detection in concrete tunnels using a Gaborfilter invariant to rotationrdquo Sensors vol 17 no 7 p 1670 2017
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Journal of Sensors 11
Bolt holeConcrete spalling
Segment seam
7
6 5 4
32
1
Figure 14 Recognition result of bolt holes segment seams andconcrete spalling patches in tunnel subinterval of 175-200 m
could not be detected are very small and belong to the normalcategory which will not affect the performance and reliabilityof a tunnel
4 Conclusion
In this paper a new method used for concrete spallingdetection in metro tunnel from point cloud based on theroughness descriptor is proposed Firstly the point cloudacquired by mobile laser scanning system needs to eliminateoutlier points originating from ancillary facilities attachedto shield tunnel wall based on the residual error filteringalgorithmThen a roughness descriptor for the metro tunnelsurface is designed to extract the rough areas on the tunnelsurface including bolt holes segment seams and spallingpatches Finally rough area classification is performed onthe identified rough areas to accurately separate the segmentseams and bolt holes from rough areas so that the concretespalling patches are left A section of metro tunnel intervalabout 250m in Shanghai is selected to verify the validityof the proposed method and seven concrete spalling areasare detected which are identified as surface defects in metrotunnel This could be helpful for tunnel maintenance andoperation safety Compared with previous studies the con-cept of roughness descriptor is proposed to detect concretespalling which is suitable for not only flat concrete surfacesbut also nonplanar concrete surfaces and at the same timeoffer the guidance for optimal scanning parameter selection
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This research is supported by the National Science Foun-dation of China (no 41671451) the National Science andTechnologyMajor Program (2016YFB0502104) and the Fun-damental Research Funds for the Central Universities ofChina The authors would like to express appreciation tocolleagues in our laboratory for their valuable commentshelp
References
[1] T Asakura and Y Kojima ldquoTunnel maintenance in JapanrdquoTunnelling and Underground Space Technology vol 18 no 2-3pp 161ndash169 2003
[2] Y Yuan Y Bai and J Liu ldquoAssessment service state of tunnelstructurerdquo Tunnelling and Underground Space Technology vol27 no 1 pp 72ndash85 2012
[3] F Sandrone and V Labiouse ldquoIdentification and analysisof Swiss National Road tunnels pathologiesrdquo Tunnelling andUnderground Space Technology vol 26 no 2 pp 374ndash390 2011
[4] Portland Cement Association (PCA) Concrete slab surfacedefects Causes Prevention And Repair Portland CementSkokie IL USA 2001
[5] N Delatte S Chen N Maini et al ldquoApplication of non-destructive evaluation to subway tunnel systemsrdquo Transporta-tion Research Record vol 1845 no 3 pp 127ndash135 2003
[6] H Russell and J Gilmore ldquoInspection policy and proceduresfor rail transit tunnels and underground structuresrdquo TransitCooperative Research Program Synthesis of Transit Practice 1997
[7] Y Yuan X Jiang and Q Ai ldquoProbabilistic assessment forconcrete spalling in tunnel structuresrdquo ASCE-ASME Journalof Risk and Uncertainty in Engineering Systems Part A CivilEngineering vol 3 no 4 2017
[8] M-K Kim H Sohn and C-C Chang ldquoLocalization andquantification of concrete spalling defects using terrestrial laserscanningrdquo Journal of Computing in Civil Engineering vol 29 no6 2015
[9] Z Zhu and I Bilakis ldquoDetecting air pockets for architecturalconcrete quality assessment using visual sensingrdquo ElectronicJournal of Information Technology in Construction vol 13 pp86ndash102 2008
[10] American Concrete Institute ACI Manual of Concrete Inspec-tion ACI Committee 311 SP-2(07) Detroit Mich USA 2007
[11] A M Paterson G R Dowling and D A ChamberlainldquoBuilding inspection can computer vision helprdquo Automationin Construction vol 7 no 1 pp 13ndash20 1997
[12] B Guldur Erkal and J F Hajjar ldquoLaser-based surface damagedetection and quantification using predicted surface proper-tiesrdquo Automation in Construction vol 83 pp 285ndash302 2017
[13] T Dawood Z Zhu and T Zayed ldquoMachine vision-basedmodelfor spalling detection and quantification in subway networksrdquoAutomation in Construction vol 81 pp 149ndash160 2017
[14] RMedina J Llamas J Gomez-Garcıa-Bermejo E Zalama andM Segarra ldquoCrack detection in concrete tunnels using a Gaborfilter invariant to rotationrdquo Sensors vol 17 no 7 p 1670 2017
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
12 Journal of Sensors
[15] S German I Brilakis and R Desroches ldquoRapid entropy-baseddetection and properties measurement of concrete spallingwith machine vision for post-earthquake safety assessmentsrdquoAdvanced Engineering Informatics vol 26 no 4 pp 846ndash8582012
[16] C Koch and I Brilakis ldquoPothole detection in asphalt pavementimagesrdquo Advanced Engineering Informatics vol 25 no 3 pp507ndash515 2011
[17] T C Hutchinson and Z Chen ldquoImproved image analysis forevaluating concrete damagerdquo Journal of Computing in CivilEngineering vol 20 no 3 pp 210ndash216 2006
[18] H S Park H M Lee H Adeli and I Lee ldquoA new approachfor health monitoring of structures terrestrial laser scanningrdquoComputer-Aided Civil and Infrastructure Engineering vol 22no 1 pp 19ndash30 2007
[19] M Hawarey and M O Falk ldquoUsing laser scanning technologyto measure deflections in steel columnsrdquo Iron and Steel Technol-ogy vol 1 no 3 pp 40ndash45 2004
[20] S J Gordon andDD Lichti ldquoModeling terrestrial laser scannerdata for precise structural deformation measurementrdquo Journalof Surveying Engineering vol 133 no 2 pp 72ndash80 2007
[21] G Teza A Galgaro and F Moro ldquoContactless recognition ofconcrete surface damage from laser scanning and curvaturecomputationrdquo NDT amp E International vol 42 no 4 pp 240ndash249 2009
[22] T Mizoguchi Y Koda I Iwaki et al ldquoQuantitative scalingevaluation of concrete structures based on terrestrial laserscanningrdquo Automation in Construction vol 35 pp 263ndash2742013
[23] W Liu S Chen and E Hauser ldquoLiDAR-based bridge structuredefect detectionrdquoExperimental Techniques vol 35 no 6 pp 27ndash34 2011
[24] P Tang D Huber and B Akinci ldquoCharacterization of laserscanners and algorithms for detecting flatness defects on con-crete surfacesrdquo Journal of Computing in Civil Engineering vol25 no 1 pp 31ndash42 2011
[25] J Yoon M Sagong and J S Lee ldquoDevelopment of damagedetection method on the tunnel lining from the laser scanningdatardquo in Proceedings of theWorld Tunnel Congress 2007 and 33rdITAAITES Annual General Assembly pp 1469ndash1474 2007
[26] A Martin and C Robert ldquoRandom sample consensus aparadigm for model fitting with applications to image analysisand automated cartographyrdquo Communications of the ACM vol24 no 6 pp 381ndash395 1981
[27] M Kazhdant M Bolitho and H Hoppe ldquoPoisson surfacereconstructionrdquo in Proceeding SGP rsquo06 Proceedings of the fourthEurographics symposium on Geometry processing pp 61ndash702006
[28] Y Lei Shouzheng T and S Xinyu ldquoAn algorithm of stemsurface reconstruction based on cylindrical projectionrdquo Journalof Forest Research vol 29 no 6 pp 812ndash819 2016
[29] W Dunham ldquoHeronrsquos formula for triangular areardquo in JourneythroughGeniusTheGreatTheorems ofMathematics pp 113ndash132Wiley New York NY USA 1990
[30] R O Duda and P E Hart ldquoUse of the Hough transformationto detect lines and curves in picturesrdquo Communications of theACM vol 15 no 1 pp 11ndash15 1972
[31] D Comaniciu and P Meer ldquoMean shift a robust approachtoward feature space analysisrdquo IEEE Transactions on PatternAnalysis and Machine Intelligence vol 24 no 5 pp 603ndash6192002
[32] H Hotelling ldquoAnalysis of a complex of statistical variables intoprincipal componentsrdquo Journal of Educational Psychology vol24 no 7 pp 498ndash520 1933
[33] P J Besl and N D McKay ldquoA method for registration of 3-D shapesrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 14 no 2 pp 239ndash256 1992
[34] H Xiaotong and W Jiandong ldquoSimilarity analysis of three-dimensional point cloud based on eigenvector of subspacerdquoHongwai yu Jiguang GongchengInfrared and Laser Engineeringvol 43 no 4 pp 1316ndash1321 2014
