Top Banner

Click here to load reader

An event detection method using floating car data

Jun 12, 2015

ReportDownload

Documents



This paper presents a method which detects events from floating car data (FCD). In traffic prediction, an event is one of unexpected factors which deteriorates the prediction performance. If occurrences of events are provided to prediction system beforehand, the prediction will be improved. Firstly, we confirmed contribution of event information to a traffic prediction accuracy. Then the paper will show that spatio-temporal pattern of zonal traffic attraction and generation is an intelligible trail of an event. Finally, the paper will propose an event detection and prediction method using PCA and HMM. The result shows feasibility of event prediction in high accuracy in certain condition.

  • 1. An event detection method usingfloating car data Osamu Masutani, Hirotoshi IwasakiDenso IT Laboratory, Inc. Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved.

2. SummaryBackgroundConceptMethodology- Factorization- Detection- PredictionConclusions Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved. 2/17 3. Background:Event Ex. Festivals, sports event As a destination - Attractive POI - Drivers would like to know fresh information about events.Our aspect As an obstacle - Repellent POI. - Drivers would like to know risk ofEvent congestion caused by an event. Event information - Place and time!Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved.3/17 4. Background:Avoidance of event congestion Traffic information service - Watch and avoid Congestion-aware route guidance Traffic prediction Problem - Most of prediction method based onGoogle maps stationery situation of traffic. - Event is irregular and unpredictable.Eventahead Event plan information can18:00 improve prediction accuracy. - Event-aware traffic predictionCopyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved.4/17 5. Background:Event database Manually collected database - For web, telematics services Not integrated - Most of services are dedicated to certain genre (ex. business event, leisure event) - Private event information isnt provided Fireworks in Stockholm (ex. school event) Not real-time - Re-schedule isnt always trackedCopyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved.5/17 6. Concept: Event detection method with FCD Fully automatic detection - Extracted from floating car data (FCD) Integrated - Can detect any type of event. - Private event also be detected. Real-time - Based on real-time fluctuation of FCDCopyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved. 6/17 7. Concept:Relationship between FCD and an eventAttractionAssumptionCount of attraction and generation of cars to/froman event venue indicates event-specific pattern. Stadium Definition - Zone : gridded areas - Attraction : Incoming cars to the zone - Generation : Outgoing cars from the zone Generation Event period Event specific pattern: Attraction increase before event and generation increase after eventAttractionGenerationCopyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved.7/17 8. Concept: 3D view of attraction and generationEvents Spatio-temporal density view reveals the event-specific pattern - Attraction peak (blue chunk) followed by generation peak (red chunk) Density mapping by iso-surface - More interpretable than point scatterStadiumSimilar method with crime mappingNakaya, T., Yano, K., (2008).Spatio-temporal three-dimensional mapping of crime events:visualizing spatio-temporal clusters of snatch-and-run offencesCopyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved.8/17 9. Factorization: For accurate detectionAn event venue isntidentical with a grid zone- Some sets of zones are relatedwith one event StadiumEvent specific pattern mightbe hidden in stationerytrafficEvent fluctuation- Various factors of fluctuation arecoincident in a zone- Stationary fluctuationEvent related factor should beproperly separated Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved.9/17 10. Factorization: Spatio-temporal factorizationPCA on spatio-temporal space- Factor is a pair of zone and timeseries.Factored zones are zones whichhave coincident traffic.Factored time series areseparated according to itspattern. Similar method with image factorizationOliver, N. M., Rosario, B., Pentland, A. P., (2000).A Bayesian Computer Vision System forModeling Human Interactions Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved. 10/17 11. Factorization: Factors for density of attractionCategorization of factors- Single zone / set of zones- Periodic / irregular /transitory fluctuationFactor 1 : Weekday morning peaks on central station and business areaFactor 2 : Evening peaks on night spotFactor 3 : Transitory peaks on a sports centerFactor 4 : Weekday midnight peaks on central stationFactor 5 : Transitory peak on a spot Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved. 11/17 12. Factorization: Event related factor Zones: nearby area of the stadium Time series: corresponding to the dates of baseball games Stadium Weighted combination of zones : baseball games Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved.12/17 13. Detection: Event pattern detection by HMM In-event Non-eventEvent state definition- 4 states (non, pre, in, post)- Pre- and post- event states defined as2 hours before and after the eventperiod.Pre-eventPost-eventDetection of pre-event can beregarded as prediction of event4 state HMM- Trained by pair of actual event data andFCD Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved. 13/17 14. Detection: Detection resultsEvaluation on the stadiumevents- 1.5 month, near a stadium, 17events (baseball games) Event states and raw dataResults confirms ourmethod works well- Event states detection:precision = 86%- Event occurrence detection:precision = 100%Recall = 100% Detected events Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved. 14/17 15. Prediction: Event aware traffic predictionBenefit of event information- Based on statistical prediction- Enhanced by adding a prediction pattern for event NormalEvent awareWeekdayWeekday HolidayHoliday Event Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved. 15/17 16. Prediction: Evaluation resultEvaluation- Comparison event aware and normal (baseline) prediction- 1.5 month, near a stadium, 17 events (baseball games) Stadium and trafficEvent information improves prediction accuracy- Overall accuracy of event-aware prediction outperforms 38% overnormal prediction Event period 38% Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved. 16/17 17. ConclusionsFCD attraction and generation statisticsis good clue to event occurrenceFactorization and HMM can detectrelatively large scale eventsEvent can refine traffic prediction abilityFuture works- Smaller events- Event classification Copyright (C) 2009 DENSO IT LABORATORY, INC. All Rights Reserved. 17/17