A plane-geometry model for automatic detection of visual vehicle incident

Automatic detection of vehicle incident by computer vision is one of the most important fields of video surveillance. In this paper, we propose a plane-geometry model to understand the vehicle behavior based on the visual information. The geometrical center of the vehicle-in-video object has different characters in different incidents. The vehicle objects of video are obtained by a background modeling approach, and their geometrical center is tracked by a meanshift-weight particle filter algorithm. The mathematic model of moving behavior of the tracked vehicle is for detecting the vehicle incidents including breaking, wrong-direction driving and wrong-lane driving. Finally, several traffic videos are tested, and the results indicate the proposed model is efficient, high-detection-rate and robust.