Abstract
Videos taken from a single camera are the most common media of human motions. In this paper, we present a method to reconstruct the 3D motion of a human figure from a single viewpoint video. We treat this problem as an optimization problem for finding joint angles of the human figure that fit each frame of a video. From the viewpoint of global optimization, the algorithm tracks each part of the body using probability field of each link and optical flow between the frames. We calculate image moments of the each link of the model's projection image and the frame, then optimize the joint angles through them. The experimental results show that we could estimate human motion from a image sequence.
