Motion capture has been one of the most promising technologies in character animation in recent years. However

most currently available motion capture approaches suffer from costly equipment and motion restriction caused by markers. To overcome these problems

a novel approach to extract 3D motion from video using vision technology is presented. The key issues

such as feature tracking and 3D reconstruction

are deeply studied. A model-based feature racking algorithm

which utilizes Kalman filter to predict coordinates of image features and epipolar line equation to aid tracking

is presented to track human motion with great variety. A non-coplanar nonlinear calibration model and a reconstruction approach taken uncertainty into consideration are applied to restore 3D human skeleton model. At last the experimental results and analysis of our VBHA (Video Based Human Animation) system demonstrate the feasibility and effectiveness of our approaches.