Research On Binocular Stereo Vision Technology Based On Mobile Robot

The binocular stereo vision, based on D. Marr's computational vision theory, is an efficient approach for recovering range information from left and right images.The thesis places the emphases on the research on camera calibration, stereo matching and mobile robot's obstacle avoidance based on binocular stereo vision. Based on Heikkila's camera model and Zhang's two-stage approach, this paper brings forward an improved camera calibration approach. This approach fully considers the effect of tangential distortion and gradient of the pixels, so it has a more widely application. The calibration experiment proved that this method has many characteristics such as convenience, higher precision and better robustness. At the same time, the intrinsic and extrinsic parameters of the binocular camera used in this thesis's experiments have been calibrated by this approach.Aiming at the characteristics of mobile robot's fast moving, this thesis brings forward an improved WTA stereo matching algorithm based on the analysis of the common stereo matching algorithms. Several stereo matching validations, subpixel interpolation and edge detection are adopted in stereo matching to ensure the validity of matching and to reduce the ratio of bad matching. After stereo matching, object's distance can be obtained by means of triangulation measurement principle and has a better precision.In addition, based on the distance measurement of stereo vision and combined with the sonar sensor and PSD sensor, this thesis realizes the location and avoidance of obstacle in complicated situation with some obstacle avoidance policy.The work of this thesis is the prior tasks of binocular stereo vision system of mobile robot, which would be the base of more perfect system in the future.