March 9, 2011
Title: Model-Based Human Pose Tracking
Abstract: Recovery and analysis of human pose and motion from video is a key enabling technology for myriad potential applications, such as smart mobile devices, advanced surveillance systems, and new perceptual man-machine interfaces. Despite more than decade of focused research, however, the general problem of detecting and tracking people in natural environments, from monocular observations, remains challenging. Pose estimation is especially challenging because image measurements are often insufficient to fully constrain 3D pose, and as a consequence, current techniques rely heavily on prior models of human model. This talk will describe the nature of the problem and some recent advances in modeling human motion based on kinematic data, physics-based constraints, and biomechanical considerations.