Flow lookup and biological motion perception

Optical flow in monocular video can serve as a key for recognizing and tracking the three-dimensional pose of human subjects. In comparison with prior work using silhouettes as a key for pose lookup, flow data contains richer information and in experiments can successfully track more difficult sequences. Furthermore, flow recognition is powerful enough to model human abilities in perceiving biological motion from sparse input. The experiments described herein show that a tracker using flow moment lookup can reconstruct a common biological motion (walking) from images containing only point light sources attached to the joints of the moving subject.