Simple Human Torso Tracking from Video

Stefan Weber

This report describes an implemenation of a basic vision system for 2.5D feature tracking of the human torso. The tracking system provides a motion sequence of a stick figure in image coordinates. The stick figure is a kinematic model of the upper body, including the hands, elbows, shoulders, neck, head, and lower torso. The tracking algorithm uses an assumed posture to calibrate the kinematic model. In subsequent frames, signal processing techniques are used to estimate feature locations, using the updated kinematic model to constrain the search space.