Articulated body tracking using belief propagation with disparity map

The articulated body tracking is challenging area in HCI (Human Computer Interaction) community. In this paper, we propose a method for robust articulated body tracker using dense disparity maps, which reduce the searching space, derived from stereo image sequences. To track an articulated body we first model the human body as MRF networks with tree structured graph, and belief propagation(BP) is then carried out to find the MAP(Maximum a posterior) of each body configuration. To track fast and accurately, we use dense disparity maps which roughly generated by block matching algorithm but post processing make their quality adequate for BP. Our proposed method 1) Generates dense disparity map and foreground and background are separated. Then the searching space of nodes is reduced taking foreground region of disparity maps. 2) Disparity information is used to calculate compatibility function to track more accurately than others using only two dimensional information.