TRACKING PEOPLE UNDER OCCLUSION, UNCERTAINTY AND DISCONTINUOUS MOTION
Dr Jamie Sherrah, Machine Vision Group, Department of Computer Science, Queen Mary and Westfield College, University of London

Abstract: Real-time tracking of people from single camera views is becoming increasingly important, not only in surveillance but more so in human-computer interfaces for telecommunications, entertainment and high-tech living. The main challenges to this endeavour are threefold: mutual occlusion of body parts and people, uncertainty in measurements due to noise and clutter, and motion discontinuities due to rapid body movement or limited computational resources. I will present our approach to tracking a person's head and hands from a single near-frontal view. A discrete Bayesian network that fuses visual cues, namely colour, motion and local orientation, with high-level knowledge and constraints. Through probabilistic inference, the network deduces hand positions over time without sole reliance on spatio-temporal continuity. The robustness of the tracker is demonstrated in a real-time video conferencing system that controls a pan-tilt-zoom camera in response to human gestures.

This seminar was held at the Department of Computer Science, Royal Holloway, University of London on 13 November 2000.

back