Skip to Main Content
In this paper, we propose a new markerless model-based human body motion capture algorithm. It no longer requires the foreground segmentation as an essential introductory step. The algorithm combines the human body model with pose parameters and the images from multiple cameras, and conducts segmentation and motion capture simultaneously via active contours and level set. In the process of curve evolution, the curve drives the human model close to the human real pose in each camera. On the other hand, when the human model is superposed with the human real pose, the curve will be balanced at the best position based on the prior shape segmentation. Promising results on real images demonstrate the potentials of the presented method.