Tracking of human movement is essential in diagnosis, follow up and quality assurance in treatment concerning the human
locomotor apparatus. The high accuracy required has led to the dominance of measurement systems that rely on instrumentation of the patient, often in the form of small reflective spheres (markers) attached to the skin. In some situations, the use of markers and marker-based measurements systems becomes an obstacle. Moreover, marker-based systems throw away a lot of information available in high-quality video images, retaining only the coordinates of the markers. Better tracking may be attained by utilizing this information.

We have attacked the problem of tracking human motion from video sequences from two sides. One approach is model-based, meaning that models of the shape of the body segments, and models of the joints are used. Features of the model, such as edges, are matched to edges in the image sequence. The other approach is by tracking patches of skin through the image sequence. Image analysis, rigid body kinematics and state estimation using Kalman- and particle filtering are central parts of the method development.

Publications

• Björn Holmberg PhD thesis Estimating Human Limb Motion Using Skin Texture and Particle Filtering 2009
• Kjartan Halvorsen PhD thesis Model-based Methods in Motion Capture 2002