Capturing Human Body Motion from Video for Perceptual Interfaces by Sequential Variational MAP

In this paper, we propose a novel sequential variational maximum a posteriori (MAP) algorithm to recover the articulated human body motion from video for perceptual interfaces. Most probabilistic methods for visual tracking adopt the mean values of the motion posteriors as the estimate. This is due to the general difficulty of the global optimization involved in the MAP estimation. However, the mean estimate is confronted with the tracking failure resulted from the multi-mode motion posteriors. We show, with theoretic guarantee, that the MAP estimate could be asymptotically achieved from a probabilistic variational approach. This new algorithm, namely sequential variational MAP, could recover the human articulation more robustly. It also achieves linear complexity w.r.t. the number of body parts, which greatly relieves the curse-of-dimensionality. Our experimental results demonstrate the effectiveness and efficiency of the proposed algorithm for articulated human body tracking, and its applicability to vision based perceptual interfaces.