Object Recognition And Localization Based On Vision In Robotic Mapping

Robotic mapping has been highly active research area in robotics and AI as to its key problem for navigation and exploration in unknown environment. The main solutions to the problem are based on active sensors such as laser and radar to collect environment information, which fail in 3D detection. So computer vision makes the future research.It is studied of object recognition and localization based on a single camera. So structure from motion is applied in the system, which includes camera calibration, feature extraction, stereo match, estimation of the fundamental matrix, estimation of the essential matrix, camera motion and 3D coordination and so on. The camera is calibrated through two-steps by Zhang. Corners are extracted by SUSAN detector, correlation and relaxation are applied to the corners and produces a matched point set, which are judged by order constraint to increase the efficiency. Then fundamental matrix is estimated by Particle Swarm Optimization. Finally 3D coordination is computed.Redefinition of order-constraint and a valid algorithm is presented to get a more robust match point set. The second contribution is fundamental estimation based on PSO on the study of 8-points and M-estimator. It is proved to be available, valid and robust.