Combining vision and computer graphics for video motion capture

We describe innovative methods for extracting three-dimensional motion of humans and animals from unrestricted monocular video, using a combination of new and established computer vision and computer graphics techniques. We identity features using image processing and active contours. Active contours become anchored to model segments in specified image frames and automatically "pull" the segments to feature positions in other frames. Adjustments are subject to joint limits and may use inverse kinematics. Occluded contour points are detected using object geometry and do not participate in feature tracking. Interactive adjustments are possible at any time in the process, allowing extraction of complicated movements regardless of background, camera movement, or feature clarity.