A Bayesian approach to activity detection in video using multi-frame correlation filters
Abhijit Mahalanobis, Robert Stanfill, Kenny Chen · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2011
Multi-frame correlation filters have been recently reported in the literature for the detection of moving objects. Introduced by Kerekes and Kumar [5], this technique uses a motion model to accumulate evidence over time in a Bayesian framework to improve the receiver operating characteristic (ROC) curve. In this paper, we generalize the approach to not only detect objects, but also their activities by using separate motion models to represent each activity. We also discuss results of preliminary simulations using publicly released aerial data set to illustrate the concept.