Analysis of 3D human motion in a computer vision- based motion capture system

This paper introduces an analyzing method of a human motion in a real-time 3D computer vision system called MIMIC. An actor acts without any attached marker in surrounded place with three CCD cameras and is wearing normal clothes freely. We extract 2D information of the root and the end-effectors of an actor such as a head, hands, and feet from inputted images, and generate 3D position of the root and these end-effectors. To construct an entire virtual human model, we also reconstruct the positions of intermediate joints using anatomically constrained inverse kinematics algorithm.