Abnormal behavior detection and behavior matching for networked cameras

In this work we consider two problems for video surveillance applications: (a) abnormal behavior detection and (b) behavior matching across cameras. We propose busy-idle rates, meaningful and easy to compute features of foreground objects, to characterize the behavior profile of a given pixel. We use these features to model the typical behavior that is observed in training sequences. Using a small number of samples for each pixel we generate behavior clusters, wherein pixels with similar behavior profiles fall into the same cluster. We then generate probabilistic models corresponding to behavior clusters, and use these models to perform abnormal behavior detection. We next show geometry independence properties of busy-idle rates. Simply stated, a set of objects observed by multiple cameras, under certain conditions, generate similar busy-idle statistics in each camera, and this holds true regardless of the camera orientation with respect to the scene and regardless of the zoom levels. We demonstrate this result via real world camera networks. Based on the premise of geometry independence, we use busy-idle rates and bring a novel approach to behavior matching problems, where the segments of image frame that exhibit similar behavior profiles are matched across cameras. This novel approach deviates from geometry based methods, and greatly simplifies the behavior matching problem.