Event recognition with fragmented object tracks

Chan, M., Hoogs, A., Sun, Z., Schmiederer, J., Bhotika, R., and Doretto, G.
Event recognition with fragmented object tracks
Technical Report 2006GRC038, GE Global research, 2006. Visualization and Computer Vision Laboratory

Download

(unavailable)

Abstract

Complete and accurate video tracking is very difficult to achieve in practice due to occlusions, traffic, shadows and appearance changes. In this paper, we study the feasibility of event recognition when object tracks are fragmented at various levels. By changing the lock score threshold controlling track termination, different levels of track fragmentation are generated. In addition, the data contains lengthy occlusions of objects involved in the events. The effect on event recognition is revealed by examining the event model match score as a function of lock score threshold. Using a Dynamic Bayes network to model events, it is shown that event recognition improves with greater track fragmentation, assuming fragmented tracks for the same object are linked together. This is counter-intuitive, as it implies that event recognition works better as tracking degrades. The experiments also show that the model is capable of handling long tracking gaps. The study is conducted on busy scenes of airplane servicing activities, with occlusions lasting hundreds of frames, object appearance changes and significant clutter traffic.

BibTeX

@TechReport{chanHSSBD06tr,
  Title                    = {Event recognition with fragmented object tracks},
  Author                   = {Chan, M. and Hoogs, A. and Sun, Z. and Schmiederer, J. and Bhotika, R. and Doretto, G.},
  Institution              = {GE Global research},
  Year                     = {2006},
  Address                  = {Niskayuna, NY, USA},
  Month                    = jan,
  Note                     = {Visualization and Computer Vision Laboratory},
  Number                   = {2006GRC038},
  Abstract                 = {Complete and accurate video tracking is very difficult to achieve in practice due to occlusions, traffic, shadows and appearance changes. In this paper, we study the feasibility of event recognition when object tracks are fragmented at various levels. By changing the lock score threshold controlling track termination, different levels of track fragmentation are generated. In addition, the data contains lengthy occlusions of objects involved in the events. The effect on event recognition is revealed by examining the event model match score as a function of lock score threshold. Using a Dynamic Bayes network to model events, it is shown that event recognition improves with greater track fragmentation, assuming fragmented tracks for the same object are linked together. This is counter-intuitive, as it implies that event recognition works better as tracking degrades. The experiments also show that the model is capable of handling long tracking gaps. The study is conducted on busy scenes of airplane servicing activities, with occlusions lasting hundreds of frames, object appearance changes and significant clutter traffic.},
  Bib2html_pubtype         = {Tech Reports},
  Bib2html_rescat          = {Video Surveillance, Event Recognition},
  File                     = {chanHSSBD06tr.pdf:doretto\\report\\chanHSSBD06tr.pdf:PDF;chanHSSBD06tr.pdf:doretto\\report\\chanHSSBD06tr.pdf:PDF},
  Keywords                 = {video surveillance, event recognition, semantics, dynamic bayesian network, video analysis},
  Owner                    = {doretto},
  Timestamp                = {2006.11.29}
}