Wide-area recognition using Hybrid Motion Stereo

In this paper, a wide-area recognition technique for mobile robots, which proves to be specifically effective with biped robots, using Hybrid Motion Stereo is presented. The proposed technique uses several cameras and divides the field of view into two groups, namely, areas visible in more than two cameras (MCVA : Multiple Camera Visible Area) and areas visible in only a single camera (SCVA : Single Camera Visible Area). Positional informations contained in MCVA can be computed constantly by stereo vision, while those contained in SCVA require movement of the camera to be computed by motion stereo. Estimation of the movement of cameras can be obtained by calculating the movement of immobile objects in MCVA. The measure of precision in computation by motion stereo depends on the magnitude of the movement of points in the image. Thus, high precision in SCVA can be obtained by lateral or longitudinal motions of the cameras. One such motion can be generated with biped robots for its lateral cyclic motion caused by the movement of the center of gravity for stability retainment. The authors have implemented these methods into the humanoid robot HRP-2 and evaluated the effectivity of the technique.