Human motion from active contours

We describe an approach for extracting three-dimensional articulated motion from unrestricted monocular video sequences. We combine feature extraction methods based on active contours with interactive adjustment. An articulated model is interactively aligned with the image in selected anchor frames. Active contour points are anchored to model segments in these frames. Occluded points are detected using object geometry and do not participate in edge tracking. Model joints are automatically adjusted in other frames to align with active contour points. The combination of interactive and automatic adjustment allows extraction of arbitrarily complex movements