Perceptual system of networked partner robots

This paper discusses how to extract human motion patterns of networked partner robots. Typically, human motions can be classified into body motions and gesture motions, and are extracted by human tracking and gesture recognition, respectively. In order to improve the performance of these motion recogntion methods, we use both of CCD camera image and distance information of a laser range finder. The proposed method for the human tracking is composed of differential extraction, particle velocity estimation, extraction of facial direction, and adaptive color matching, and 3D human modeling of laser range finder. Finally, we discuss the effectiveness of the the proposed method through several experimental results.