Recovery of Vehicle Trajectories from Tracking Data for Analysis Purposes Dalia Tiešytė, Christian S. Jensen {dalia,csj}@cs.aau.dk
ITS 2007, Aalborg, Denmark
The TransDB Project • TransDB: www.cs.aau.dk/TransDB)
Vehicle tracking with focus on scheduled journeys Historical data analysis Prediction of travel times and improvement of schedules
Historical positions
Statistical analysis
Travel patterns
Tracking GPS data
Tracking
Prediction
PerTrans, 23 March 2007, White Plains, NY, USA
Scheduling
2
The Tracking System Example Historical data repository
GPS Position Position
2 18:05 AAU Busterm
Central server
Position-related information
Threshold based tracking
Process positionrelated data PerTrans, 23 March 2007, White Plains, NY, USA
3
Roadmap • Background
Tracking, prediction, and historical data analysis
• Problem statement
The trajectory function The most optimal trajectory
• The recovery algorithm • Analysis and experimental results • Discussion
PerTrans, 23 March 2007, White Plains, NY, USA
4
Background • Tracking algorithms
Shared movement prediction [Čivilis et al., Wolfson et al., etc.]
pos (t ) = f ( posup , tup ) • Movement prediction functions
Linear movement function. Deviation from the schedule. Advanced prediction (e.g., neural networks).
• Recovery of vehicle trajectories
Assume linear movement inbetween updated points.
PerTrans, 23 March 2007, White Plains, NY, USA
5
Position ThresholdBased Tracking • Share the vehicle’s movement function between the server and the vehicle. • Update when the GPS position deviates by threshold thr from the server’s predicted position. • Position accuracy guarantees (thr).
PerTrans, 23 March 2007, White Plains, NY, USA
6
Problem Statement • Available data:
A sequence of position updates with accuracy guarantees. Movement prediction function.
• Recover vehicle’s trajectory that:
Is as close as possible to the actual trajectory. Preserves accuracy guarantees.
PerTrans, 23 March 2007, White Plains, NY, USA
7
Recovery of Trajectories v = vmax
The updated position is later than expected
The updated position is later than expected
Assume the max. speed v max before the update
Assume the min. speed (0) before the update
PerTrans, 23 March 2007, White Plains, NY, USA
8
Analytical Study • The recovered (expected) trajectory fexp
Preserves accuracy guarantees thr:
| f exp (t) – f act(t)| ≤ thr
Is at least as close to the actual trajectory fact as the initial prediction.
| f exp (t) – f act(t)| ≤ | f pred (t) – f act(t)|
Is the “best” recovery with the given data.
′ (| f exp (t) – f act(t)| ≤ | f exp ′ (t) – f act(t)|) ∀f exp PerTrans, 23 March 2007, White Plains, NY, USA
9
Experimental Results
PerTrans, 23 March 2007, White Plains, NY, USA
10
Discussion • The proposed techniques recover closetoactual trajectories of vehicles.
The “best” with the given data.
• They may be used for…
pattern mining, better predictions of future trajectories, better scheduling,…
• Possible future research directions
Use all available constraints (e.g., speed limits). Investigate, how the recovered trajectories can be used by similarity functions, clustering algorithms, etc. PerTrans, 23 March 2007, White Plains, NY, USA
11
Acknowledgements •
Nordjyllands Trafikselskab (NT) www.nordjyllandstrafikselskab.dk
• Telenor Connect A/S (TNC Connect) www.telenorconnect.com
• Forskningsstyrelsen (Danish research agency) www.forsk.dk
• Aalborg University and Daisy group daisy.aau.dk
PerTrans, 23 March 2007, White Plains, NY, USA
12
Thank you!
Questions
?
PerTrans, 23 March 2007, White Plains, NY, USA
13