CVPR2015読み会 "Joint Tracking and Segmentation of Multiple Targets"

1. CVPR2015 Joint Tracking and Segmentation of Multiple Targets @yuukicammy 2015.7.20

2. Joint Tracking and Segmentation of Multiple Targets Anton Milan, Laura Leal-Taixe, Konrad Schindler, Ian Reid Multiple Object Tracking Benchmark (MOTChallenge) http://motchallenge.net/ MOTChallenge 13Tracker (2015.7.19) 11Anonymous submission(ICCV2015 ) CVPR2015 1

3. CRF state-of-the-arttracking-by-detection10% ID CVPR2015 2

4. Contribution CRF CVPR2015 3

5. F boundingbox(BBox) CVPR2015 4

6. Tracking-by-Detection ID 01 CVPR2015 5

7. SegmentationandTracking Voxel(volumepixel) () () CVPR2015 6

8. CVPR2015 7 boundingbox(BBox) BBox (ID) CRF(conditionalrandomfield)

9. CVPR2015 8 TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF

10. CRF CVPR2015 9 :d :s :vw : : () :(

11. BBox(x,y,w,h)4D CVPR2015 10 (, ):(Bbox) (, ):Bbox

12. HOG(HistogramofOrientedGradients) HOF(HistogramsofFlow)SVM http://www.milanton.de/data.html CVPR2015 11 TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF

13. DP[*]Context-free Tracker[**] () N CVPR2015 12 TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF [*]H. Pirsiavash, D. Ramanan, and C. C. Fowlkes. Globally optimal greedy algorithms for tracking a variable number of objects. In CVPR 2011 [**] J. Henriques, R. Caseiro, P. Martins, and J. Batista. Exploiting the circulant structure of tracking-by-detection with kernels. In ECCV 2012

14. TSP(TemporalSuperPixel) Chang[*] CVPR2015 13 [*] J.Chang,D.Wei,andJ.W.FisherIII.A video representation using temporal superpixels. In CVPR 2013 TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF

15. / CVPR2015 14 () BBox () Lab k-means5 5% TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF

16. CVPR2015 15 /SVM Lab i score:SVM TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF

18. CVPR2015 17 j Di:Bbox Tj:j () TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF

19. CVPR2015 18 Sij TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF

21. CVPR2015 20 () () () TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF

22. CVPR2015 21 () Lab TSP TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF

24. CVPR2015 23 () BBox BBox Imagelikelihood TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF

25. CVPR2015 24 CRF 12 TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF

26. CVPR2015 25 [*] TSP(TemporalSuperPixel) HOG&HOF+SVM SVM/ CRF [*] A. Milan, K. Schindler, and S. Roth. Detection and trajectory-level exclusion in multiple object tracking. In CVPR 2013.

27. MatlabC++ matlabliblinearGSL Anton TSP 2.7GHz,1Core 0.2fps CVPR2015 26

28. -- CVPR2015 27 MOTChallenge Dataset:Pets09(S2L1,S2L2,S2L3,S1L1-2,S1L2-1),TUDS [*] TA(MOTA):Thismeasurecombinesthreeerrorsources:falsepositives,missed targetsandidentityswitches TP(MOTP): Themisalignmentbetweentheannotatedandthepredicted boundingboxes MT(Mostlytrackedtargets):Theratioofground-truthtrajectoriesthatare coveredbyatrackhypothesisforatleast80%oftheirrespectivelifespan ML(Mostlylosttargets):Theratioofground-truthtrajectoriesthatare coveredbyatrackhypothesisforatmost20%oftheirrespectivelifespan FMThetotalnumberoftimesatrajectoryisfragmented(i.e.interrupted duringtracking) [*]Bernardin, K.&Stiefelhagen,R. EvaluatingMultipleObjectTrackingPerformance:The CLEARMOTMetrics. ImageandVideoProcessing, 2008(1):1-10, 2008.

29. -- CVPR2015 28

30. - - CVPR2015 29 http://motchallenge.net/results/2D_MOT_2015/

31. () CRF CRFBboxTemporalSuperPixel(TPS) MOTChallenge() CVPR2015 30

CVPR2015読み会 "Joint Tracking and Segmentation of Multiple Targets"

Engineering