Shared latent dynamical model for human tracking from videos

Abstract

Of different learning-based methods in human tracking, many state-of-the-art approaches have been dedicated to reduce the dimensionality of the pose state space in order to avoid complex searching in a high dimensional state space. Seldom research on human tracking refers shared latent model. In this paper, We propose a method of shared latent dynamical model (SLDM) for human tracking from monocular images. The shared latent variables can be determined easily if state vectors and observation vectors are statistically independent.With a SLDM prior over state space and observation space, our approach can be integrated into a Bayesian tracking framework of Condensation, and further a scheme of variance feedback is designed to avoid mis-tracking. Experiments using simulations and real images demonstrate this human tracking method is very efficient and promising. © Springer-Verlag Berlin Heidelberg 2007.