Video based motion capture in environments with non-stationary background

Several methods for capturing motion data from single video have been reported in computer vision literature, and most of them deal with stationary background. The problem becomes more complex and challenging in a moving scene where traditional background subtraction algorithms often fail. We require robust algorithms for marker-less tracking of human body's movements and for extracting motion information from them. This paper reviews recent research work done in the area of video based 3D motion capture through marker-less tracking, learning and detection algorithms, and identifies their usefulness and limitations. The paper then proposes a novel framework based on state-of-the-art methods for object detection and pose estimation for obtaining the 3D joint positions of a tracked human model in a single view video stream. Experimental results are presented to show the effectiveness of the proposed algorithm in capturing 3D motion information.