抄録
Marker-less Motion Capture of people is a central issue in humanoid robotics for both motion learning using the mimetic scheme and activity recognition to work alongside humans. Mobile robots are expected to adapt to a wide range of situations in the future, which requires tracking algorithms to be as general as possible. Particle filter based motion capture has been used in the literature to track the motion of a subject in color images. However, the complexity of the problem, due to the high number of degrees of freedom, makes particle filters computationally expensive. This paper proposes to constrain several degrees of freedom using image processing techniques and inverse kinematics. The resulting algorithm was tested in real-time situation and achieved robust tracking of a person's upper body.
