This project will pursue the research and development of a system to track human motions in three dimensions from single camera video input. A limited ability to sense the activities of humans in their native environment has long inhibited the application of computer technology outside strictly controlled environments. Expanding the range of what computers can learn through simple video input concerning events in their surroundings has the potential to vastly increase the number of applications to real-world situations.
The project will yield both research insights and new curricular initiatives. Current systems for tracking human motion either limit the domain in some manner (tracking only certain types of motion, or in fewer than three dimensions) or else lack reliability, losing track of their subjects over time. The research focus of this project therefore aims to build a general system with improved long-term reliability by incorporating novel sources of feedback, including retrieval of known poses from silhouettes, and graph-based techniques for segmentation and body-part identification. The mechanism will embody a flexible Bayesian framework, admitting the inclusion of additional tracking guidance components as they are developed. Once built, the tracking system will offer opportunities for curricular deployment, particularly in exercise and sport studies, where it may augment the tools available for the study of human motion and biomechanics. The work will enhance opportunities in areas utilizing motion analysis in their fields.
