Probabilistic State Space Decomposition for Human Motion Capture

Model-based approaches to tracking of articulated objects, such as a human, have a high computational overhead due to the high dimensionality of the state space. In this paper, we present an approach to human motion capture (HMC) that mitigates the problem by performing a probabilistic decomposition of the state space. We achieve this by defining a conditional likelihood for each limb in the articulated human model as opposed to an overall likelihood. The conditional likelihoods are fused by making certain conditional independence assumptions inherent in the human body. Furthermore, we extend the popular stochastic search methods for HMC to make use of the decomposition. We demonstrate with Human Eva I and II datasets that our approach is capable of tracking more accurately than the state-of-the-art systems using only a small fraction of the computational resources.