3D Human Body Tracking in Unconstrained Scenes

  • Authors:

  • Chengbin Zeng;Huadong Ma;Anlong Ming;Xiaobo Zhang

  • Affiliations:

  • Beijing Key Laboratory of Intelligent Telecommunications Software and Multimedia, Beijing University of Posts and Telecommunications, Beijing, China 100876;Beijing Key Laboratory of Intelligent Telecommunications Software and Multimedia, Beijing University of Posts and Telecommunications, Beijing, China 100876;Beijing Key Laboratory of Intelligent Telecommunications Software and Multimedia, Beijing University of Posts and Telecommunications, Beijing, China 100876;Beijing Key Laboratory of Intelligent Telecommunications Software and Multimedia, Beijing University of Posts and Telecommunications, Beijing, China 100876

  • Venue:
  • PCM '09 Proceedings of the 10th Pacific Rim Conference on Multimedia: Advances in Multimedia Information Processing
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

The 3D human body tracking from videos in unconstrained scenes is a challenging problem and has widespread applications. In this paper, we introduce a novel framework that incorporates the graph-based human limbs detection into the articulated Bayesian tracking. The 3D human body model with a hierarchical tree structure can describe human's movement by setting relevant parameters. Particle filter, which is the optimal Bayesian estimation, is used to predict the state of the 3D human pose. In order to compute the likelihood of particles, the pictorial structure model is introduced to detect the human body limbs from monocular uncalibrated images. Then the detected articulated body limbs are matched with each particle using shape contexts. Thus the 3D pose is recovered using a weighted sum of matching costs of all particles. Experimental results show our algorithm can accurately track the walking poses on very long video sequences.