The project addresses the issues of tracking and modeling of human motion. Robust tracking algorithms will be developed that do not require manual initialization and are capable of tracking over many frames without failure. These trackers will then be applied to archival video footage, to produce a rich corpus of historically-significant motion data. A combination of manual labeling, state invariant detection, and temporal co-occurrence analysis will be utilized to parse motions into sequences of movements. The generated library of learned motions can then be applied to a diverse set of applications, including robot control and computer animation. The education part of the project calls for development of a curriculum for figure tracking and motion modeling that is self-contained and accessible to students and researchers in the areas of computer vision, robotics, computer graphics, and biomechanics. An outreach program will be initiated that will target the machine learning and computer animation communities.
