A method to automatically identify road centerlines from georeferenced smartphone data

Post on 13-Feb-2016

33 views 1 download

Tags:

description

A method to automatically identify road centerlines from georeferenced smartphone data. XIV Brazilian Symposium on GeoInformatics (GEOINFO 2013). George H. R. Costa, Fabiano Baldo. {dcc6ghrc, baldo}@joinville.udesc.br. 25/11/2013. Agenda. Introduction Objective Related work - PowerPoint PPT Presentation

transcript

25/11/2013

A method to automatically identify road centerlines fromgeoreferenced smartphone data

XIV Brazilian Symposium on GeoInformatics (GEOINFO 2013)

George H. R. Costa, Fabiano Baldo{dcc6ghrc, baldo}@joinville.udesc.br

2

Agenda Introduction Objective Related work Proposed method Tests and Results Conclusion and Future work

3

Introduction Digital road maps have gained fundamental role in

population’s daily life Navigation systems etc.

It is essential that maps reflect reality as well as possible Generated from accurate data; Periodic updates.

Possible source of data: GPS traces

4

Introduction By combining many traces it is possible to generate

maps

Example: OpenStreetMap Users use uploaded traces to create/update maps However, all map editing is done manually

Automatic solutions would be more effective Could allow maps to be updated faster Feasible: [Brüntrup et.al. 2005] and [Cao and Krumm

2009] also support this idea

Challenges How to obtain the

data needed to generate maps? Smartphones

Contain many sensors, including a GPS receiver

Represent half of the Brazilian cellphone market [GFK 2013] 5

Source: Garmin

Challenges To create road maps it

is necessary to find the roads’ centerlines

How to analyze the traces to identify road centerlines? Approximated result

Evolutive algorithm6

Source: author

7

Objective Therefore, the objective of this work is to:

Propose a method to identify road centerlines using an evolutive algorithm in orderto generate and update road maps

8

Related work Characteristics gathered from other works:

Independence from initial maps [Brüntrup et.al. 2005; Cao and Krumm 2009; Jang et.al. 2010]

Usage of heuristics to remove noise from the traces [Brüntrup et.al. 2005; Cao and Krumm 2009; Zhang et.al. 2010; Niu et.al. 2011]

Characteristic introduced by this work: Traces’ date of recording is taken into account to

generate up-to-date maps

Data source

9

Source: author

Preprocessing Reduces noise; saves all traces to database

10

Source: author

11

Road centerlines

1. Query database to get all traces ordered by date and accuracyi. Most recent firstii. Most accurate first

Source:author

12

Road centerlines

2. For each point k of each trace j ():

1. Identify nearby points All points that intersect a buffer around

Source:author

13

Road centerlines

2. Points with a direction of movement different than are discarded set

3. How to analyze the set to find the road centerline?

Source:author

14

Road centerlines It is assumed that it is only possible to find an

approximated solution

Road centerline = weighted combination between: Date of recording; Accuracy; Distance from a candidate solution to all points

selected ( set).

Chosen algorithm: evolutive algorithm

15

Road centerlines

: candidate solution : set of selected points : influence of time (date of recording) : influence of accuracy : influence of distance

𝐹𝐼𝑇𝑁𝐸𝑆𝑆 (𝐶𝑥 )=∑𝑖=1

𝑛 ′ ′

𝐼𝑇 (𝑆𝑖 ) ∙𝑀𝑇 +𝐼𝐴 (𝑆𝑖 ) ∙𝑀𝐴+𝐼𝐷 (𝐶 𝑥 ,𝑆 𝑖 ) ∙𝑀𝐷

16

Road centerlines

, , : Multiply the value of the corresponding influence to prioritize desired characteristics

𝐹𝐼𝑇𝑁𝐸𝑆𝑆 (𝐶𝑥 )=∑𝑖=1

𝑛 ′ ′

𝐼𝑇 (𝑆𝑖 ) ∙𝑀𝑇 +𝐼𝐴 (𝑆𝑖 ) ∙𝑀𝐴+𝐼𝐷 (𝐶 𝑥 ,𝑆 𝑖 ) ∙𝑀𝐷

17

Road centerlines

Recent traces: weight closer to 1 Older traces: weight closer to 0

𝐹𝐼𝑇𝑁𝐸𝑆𝑆 (𝐶𝑥 )=∑𝑖=1

𝑛 ′ ′

𝐼𝑇 (𝑆𝑖 ) ∙𝑀𝑇 +𝐼𝐴 (𝑆𝑖 ) ∙𝑀𝐴+𝐼𝐷 (𝐶 𝑥 ,𝑆 𝑖 ) ∙𝑀𝐷

𝐼𝑇 (𝑆𝑖)=𝑡𝑚𝑎𝑥−|𝑇 (𝑆 𝑖 ,𝑆𝑟 )|

𝑡𝑚𝑎𝑥

Influence of Time

18

Road centerlines𝐹𝐼𝑇𝑁𝐸𝑆𝑆 (𝐶𝑥 )=∑

𝑖=1

𝑛 ′ ′

𝐼𝑇 (𝑆𝑖 ) ∙𝑀𝑇 +𝐼𝐴 (𝑆𝑖 ) ∙𝑀𝐴+𝐼𝐷 (𝐶 𝑥 ,𝑆 𝑖 ) ∙𝑀𝐷

𝐼𝐴 (𝑆𝑖 )=𝛼 𝐴 (𝑆 𝑖 )2+(−1−𝛼 ∙𝑎𝑚𝑎𝑥

2

𝑎𝑚𝑎𝑥) 𝐴 (𝑆𝑖 )+1

𝛼=−𝑎𝑙𝑖𝑚−𝑎𝑚𝑎𝑥 (𝑎𝑙𝑖𝑚𝑉−1)𝑎𝑚𝑎𝑥 𝑎𝑙𝑖𝑚 (𝑎𝑚𝑎𝑥−𝑎𝑙𝑖𝑚)

Influence of Accuracy

Wei

ght

Accuracy

19

Road centerlines𝐹𝐼𝑇𝑁𝐸𝑆𝑆 (𝐶𝑥 )=∑

𝑖=1

𝑛 ′ ′

𝐼𝑇 (𝑆𝑖 ) ∙𝑀𝑇 +𝐼𝐴 (𝑆𝑖 ) ∙𝑀𝐴+𝐼𝐷 (𝐶 𝑥 ,𝑆 𝑖 ) ∙𝑀𝐷

𝐼𝐷 (𝐶𝑥 ,𝑆𝑖 )=𝛽𝐷 (𝐶𝑥 ,𝑆𝑖 )2+(−1− 𝛽 ∙𝑑𝑚𝑎𝑥

2

𝑑𝑚𝑎𝑥)𝐷 (𝐶𝑥 ,𝑆𝑖)+1

𝛽=−𝑑𝑙𝑖𝑚−𝑑𝑚𝑎𝑥 (𝑑𝑙𝑖𝑚𝑉−1)𝑑𝑚𝑎𝑥𝑑𝑙𝑖𝑚 (𝑑𝑚𝑎𝑥−𝑑𝑙𝑖𝑚)

Influence of Distance

Wei

ght

Distance

20

Road centerlines𝐹𝐼𝑇𝑁𝐸𝑆𝑆 (𝐶𝑥 )=∑

𝑖=1

𝑛 ′ ′

𝐼𝑇 (𝑆𝑖 ) ∙𝑀𝑇 +𝐼𝐴 (𝑆𝑖 ) ∙𝑀𝐴+𝐼𝐷 (𝐶 𝑥 ,𝑆 𝑖 ) ∙𝑀𝐷

Source:author

𝐹𝐼𝑇𝑁𝐸𝑆𝑆 (𝐶𝑥 )=∑𝑖=1

𝑛 ′ ′

𝐼𝑇 (𝑆𝑖 ) ∙𝑀𝑇 +𝐼𝐴 (𝑆𝑖 ) ∙𝑀𝐴+𝐼𝐷 (𝐶 𝑥 ,𝑆 𝑖 ) ∙𝑀𝐷Closer to highest

concentration of points: smallest overall distance

Closer to points high better accuracy

𝐹𝐼𝑇𝑁𝐸𝑆𝑆 (𝐶𝑥 )=∑𝑖=1

𝑛 ′ ′

𝐼𝑇 (𝑆𝑖 ) ∙𝑀𝑇 +𝐼𝐴 (𝑆𝑖 ) ∙𝑀𝐴+𝐼𝐷 (𝐶 𝑥 ,𝑆 𝑖 ) ∙𝑀𝐷

21

Road centerlines

3. Evolutive algorithm 60 generations 20 candidate solutions per generation Elitism: 2 best candidate solutions are preserved to

the next generation

Source: author

Evolutive algorithm loop:

22

Road centerlines Evolutive algorithm finds centerline close to Next step: repeat process for If has already been used, skip to the next point

Source:author

23

Results Implemented in Python DB: PostgreSQL + PostGIS

Data collected between 27/01/2013 e 15/06/2013 4237 traces 966698 points

24

Results Tests: comparison between

Proposed method’s results Satellite images

Google Earth

Executed on places with complex road structures

25

Tests (1)

Roads intersect

Source: Google Earth / author

Satellite image

26

Tests (1)Points collected (filtered)

Source: Google Earth / author

27

Tests (1)

Way centerline

Direction of movement differentiates traces

It is possible to improve filtering...

Final result

Source: Google Earth / author

28

Tests (2)

Roads with different direction of movement

Roads with same direction of movement

Satellite image

Source: Google Earth / author

29

Tests (2)Points collected (filtered)

Source: Google Earth / author

30

Tests (2)It is possible to improve

filtering...

Direction of movement differentiates traces

Final result

It is possible to improve parameters...

Source: Google Earth / author

31

Results Small difference between the satellite images and

the method’s results Average distance (100 points): 2.95 meters

Cannot affirm which one is more accurate Certain questions cannot be controlled

Ex.: satellite images might be somewhat out of position

32

Conclusion Different from similar methods because:

Takes into consideration the influence of the traces’ date of recording;

Collects data using smartphones; Finds centerlines using evolutive algorithm.

Tests showed little difference to satellite images It is still possible to optimize parameters to achieve

better results

33

Future work Improve collected traces’ reliability

Ex.: Kalman Filter Different update policies for each region

Downtown: more data, only accept better accuracy Rural areas: less data, accept older data

Mining more information Traffic lights Pot holes

34

Bibliografia Brüntrup, R. et. al. (2005) “Incremental map generation with GPS traces”. In: Proceedings

of the 8th International IEEE Conference on Intelligent Transportation Systems. Cao, L. e Krumm, J. (2009) “From GPS traces to a routable road map”. In: Proceedings of

the 17th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems. New York, EUA: ACM Press.

Garmin (2010) “Garmin-Asus smartphones reach new markets”. <http://garmin.blogs.com/ my_weblog/2010/09/garmin-asus-around-the-globe.html> (accessed on Nov 22).

GFK (2013) “GfK TEMAX BRASIL T2 2013: Crescimento no mercado com forte influência de materiais de escritório e periféricos”. <http://www.gfk.com/br/news-and-events/press-room/press-releases/Paginas/TEMAX-BRASIL-T2-2013.aspx> (accessed on Nov 18).

Jang, S., Kim, T. e Lee, E. (2010) “Map Generation System with Lightweight GPS Trace Data”. In: International Conference on Advanced Communication Technology.

Niu, Z., Li, S. e Pousaeid, N. (2011) “Road extraction using smart phones GPS”. In: Proceedings of the 2nd International Conference on Computing for Geospatial Research & Applications. New York, EUA: ACM Press.

Zhang, L., Thiemann, F., Sester, M. (2010) “Integration of GPS traces with road map”. In: Proceedings of the 2nd International Workshop On Computational Transportation Science. San Jose, EUA. ACM Press.

25/11/2013

A method to automatically identify road centerlines fromgeoreferenced smartphone data

XIV Brazilian Symposium on GeoInformatics (GEOINFO 2013)

George H. R. Costa, Fabiano Baldo{dcc6ghrc, baldo}@joinville.udesc.br