Human motion tracking based on shape analysis

The goal of this work is to track human joints in image sequences or video fragments. In this work, we attempt to develop an algorithm to analog the perceiving process of human vision system. Based on the knowledge of human body model and shape analysis techniques used in pattern recognition, after the user selecting out edges correspond to tracked objects interactively in the first frame, the algorithm tracks human joints by using the outline and other important inner edges of a moving human figure. It does not need special markers and avoids feature searching, thus largely improves the tracking speed. Moreover, the algorithm is applied to real image sequences with very encouraging results. Experimental results indicate that, our algorithm is very robust to object moving and rotating, varying with colors, illumination and intervals between frames.