Visual Perception for Manipulation and Imitation in Humanoid Robots

This thesis deals with visual perception for manipulation and imitation in humanoid robots. In particular, real-time applicable methods for object recognition and pose estimation as well as for markerless human motion capture have been developed. As only sensor a small baseline stereo camera system (approx. human eye distance) was used. An extensive experimental evaluation has been performed on simulated as well as real image data from real-world scenarios using the humanoid robot ARMAR-III.