3D Human Tracking by Using Shared Latent Dynamical Model

Many state of the arts latent models have been investigated to learn different latent variables with Gaussian processes models (GPs). However, seldom research focus on a shared latent dynamical model (SLDM) of GPs and its application in a high dimensional nonlinear system. In this paper, we propose a shared latent dynamical model, which combines the idea of GPDM with SLS, and give its application in tracking 3D human motion from monocular videos. When tracking high dimensional states, SLDM can map state space and observation space to a shared latent space of low dimensionality with associated dynamics. During off-line training, three mappings, including dynamical mapping in latent space and mappings from the latent space to both state space and observation space, are learned. During online tracking, our approach can be integrated into a Bayesian tracking framework of Condensation, and further a scheme of variance feedback is designed to avoid failed tracking. Experiments in human motion tracking from monocular videos using simulations and real images demonstrate this human tracking method is efficient.