Markerless Motion Capture with Structure Estimation Capability

We present a markerless motion capture system able to determine the kinematic structure while measuring joint movement. In addition to volume data, we also use texture data to precisely measure the degrees of freedom that do not affect the shape, e.g., pronation/supination angles of the forearm and shank. We first obtain topology using a Reeb graph and independently build a tentative articulated-body chain model of the subject for each frame. We then extract a common optimized chain model by comparing joint angles of tentative models of all frames to identify which joints are related to describing the movement of the subject. Our system thus measures movement without prior knowledge of the structure. The system identifies the link length of objects with known structures based on measured data.