Authors: James N.K. Liu, Bernard C.S. Kan
Addresses: Department of Computing, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong. ' Department of Computing, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong
Abstract: Active map learning for autonomous robots exploring in unknown environments is an essential and vital issue for robot navigation. However, many factors limit a robot|s ability to learn accurate map models in practice. To enhance its learning capability, this paper proposed a binocular vision system, developed using IP web cameras and MATLAB image processing toolbox. It integrates knowledge from computer vision and robot navigation. Through the binocular vision system, binocular images are captured and analysed so that feature objects or shapes are identified. Together with sonar readings, this information is used in collision avoidance and map building. It is shown that the capability of active map learning is effectively realised by adopting this binocular vision system.
Keywords: mobile robots; binocular vision; colour vision; robot vision; robot navigation; sonar mapping; six points calibration; stereo matching; feature extraction; Harris feature detectors; odometry errors; obstacle avoidance; autonomous robots; map learning; robot learning; web cameras; image processing; collision avoidance.
International Journal of Intelligent Systems Technologies and Applications, 2011 Vol.10 No.1, pp.15 - 45
Available online: 25 Jan 2011 *Full-text access for editors Access for subscribers Purchase this article Comment on this article