Computer Vision and Image Understanding
| Hi-index | 0.00 |
In the domain of human motion analysis, the observed contours of the body contain rich information about the body configuration, the motion performed, the person’s identity, and even the emotional states of the person. In this paper, we introduce a framework for Bayesian tracking of the dynamic contours of the articulated human motion. We propose a factorized generative model for walking shape contour sequences that separates the dynamic deformation due to a motion, from the static variability due to the appearance of the person performing that motion. This results in an efficient tracking of gait sequence with a low-dimensional representation of body configuration and simultaneous adaptation to highly nonlinear static and dynamic shape deformations. Experimental results using the CMU Mobo, the University of Southampton (UoS), and M. Black’s walking sequence, show accurate contour tracking of a person walking by dynamic body configuration estimation on a low-dimensional manifold space and personal style estimation to fit the contour to the individual characteristics. In addition, the estimated shape style provides a good descriptors for human identification from gait.