Integrating component cues for human pose tracking

Abstract

Tracking human body pose in monocular video in the presence of image noise, imperfect foreground extraction and partial occlusion of the human body is important for many video analysis applications. Human pose tracking can be made more robust by integrating the detection of components such as face and limbs. We proposed an approach based on data-driven Markov chain Monte Carlo (DD-MCMC) where component detection results are used to generate state proposals for pose estimation and initialization. Experimental results on a realistic indoor video sequence show that the method is able to track a person during turning and sitting movements.