An Interacting Multiple Model for Total Occlusion Handling

In this paper we address the problem of vision based human tracking in a video sequence captured by a stationary camera. The proposed system consists of two parts: human detection and tracking. Due to the fact that our main object to track has a non-rigid structure, a discriminative method for detection based on a start-structured human part model is used. For the tracking, in order to handle total occlusion, a new tracking method based on a two Kalman filters Interacting Multiple Model algorithm is used. The proposed system has been tested on both synthetic and real video datasets and the experiment results demonstrate that our method is able to obtain accurate estimates of the target's state when the image of the human being tracked is occluded during unknown time periods.