Autonomous Navigation of an Intelligent Vehicle Using 1-Dimensional Optical Flow
Masaki Yamada and Kazuo Nakazawa
Machine Intelligence Laboratory, Keio University, 3-14-1 Hiyoshi, Kouhoku-ku, Yokohama 223-8522, Japan
Controlling a vehicle from hitting corridors wall was studied as a problem in visual feedback control. Control used visual information in the real world within real time. We use a 1-dimensional (1-D) optical flow calculated from intensity data obtained using a CCD linear sensor as visual information. A qualitative index indicating the proximity of the moving vehicle to the left or right wall is extracted from the optical flow. This qualitative approach enabled stable visual recognition processing without explicitly calculating distance. This method is based on general, qualitative environmental knowledge and has fewer environmental condition restraints. Because no numerical environment or vehicle models are used for control, the consideration of parameter errors, a serious problem before, is not needed.