Tracking and occlusion handling in multi-sensor networks by particle filter

In this paper we propose a multi sensor tracking method. Tracking is done independently for each view. Fusing several cues including color, edge, texture and motion constrained by structure of environment is used in a novel way and in particle filter framework. The results of individual image planes are projected to ground plane using homography relation. The similarity of projected locations with the reference model and minimum variance estimate are two key points to evaluate the location of the target. Also, we introduce a method based on two views tracking to handle occlusion. Robust statistic is used to declare an occlusion in one view. Homography relation and inter-frame transformation are the tools to cancel the occlusion. Experimental results show the robustness and accuracy of the proposed method.