Camera Self-Calibration for Unmarked Human Motion Capture System

An unmarked human motion capture system typically requires camera calibration. In this paper, we present a method for calibrating multiple cameras and set up a suitable three-dimensional coordinate system for the calibration as we want. Camera calibration is achieved by using the trajectory of a moving point. After recording the videos of point, we can get several groups of 2D coordinates respective to each camera. Then camera projective matrices are calculated with the data of 2D pixel coordinates via rank-4 factorization. The three-dimensional coordinate system which we want is determined by the position of another three points which constitute a non-isosceles right triangle. We demonstrate our calibration method by the experiments on real unmarked human motion capture system. The biggest reprojection error is less than 1.1 pixels, which is accurate enough for the motion capture.