Silhouette recognition can reconstruct the three-dimensional pose of a human subject in monocular video so long as the camera's view remains unoccluded by other objects. This paper develops a shape representation that can describe and compare partial shapes, extending the silhouette recognition technique to apply to video with occlusions. The new method operates without human intervention, and experiments demonstrate that it can reconstruct accurate three-dimensional articulated pose tracks from single-camera walking video despite occlusion of one-third to one-half of the subject.
展开▼
