A tracking method of end-effectors in a vision-based marker-free motion capture system

We describe a tracking method for analyzing 3D motions of human in a real-time vision-based marker-free motion capturing system. We capture human motions by using three synchronized color CCD cameras when an actor wearing normal clothes without any attached marker. We obtain the 2D positions of blobs such as the head, the hands, the feet and the root of an actor to use in a proposed tracking method. Then, this proposed tracking method reconstructs end-effectors having 3D coordinates from 2D blob data by using a grouping and an epipolar constraint. For correct tracking of freely moving end-effectors in real-time, we apply the spatial and temporal analysis to each 3D end-effector to distinguish right and left directions, restore a disappeared or an occluded end-effectors. Finally, we generate a 3D virtual human having 20 joints that moves like real human by using an inverse kinematics method in real-time.