4-1 Human Motion Tracking in a Single Camera with a Graph Structure in Latent Spaces

抄録

This paper presents a novel approach to track human motion with a monocular video based on particle filter. To deal with the high dimensional freedoms in human pose, a low dimensional latent space, e.g. Gaussian Process Dynamical Model (GPDM), is a typical solution. In this paper, we extend the GPDM model into a graph structure to better express the diverse dynamics in human motion, where multiple latent spaces are constructed and connected with each other.