| The omindirectional vision is a kind of technology which can obtain all vision information of 3D space larger than hemisphere. The extensive visual area of omnidirectional vision provides a kind of new method for robot to get surrounding information and to perceive its state. However, omnidirectional vision is a kind of new technology which is unripe in the field of robot navigation, and many key technologies need to be further studied. Therefore, how to use omnidirectional vision technology in robot navigation better has become a very meaningful and challenging task.Aiming at the problem that how to use omnidirectional vision sensor (ODVS) to help robot to get surrounding information more quickly and comprehensively in robot navigation, this paper has studied several key technologies about the application of omnidirectional vision technology in robot navigation. The main work can be summarized as follows:1. Study on the calibration technology of ODVS. Analyze the imaging principle of ODVS, introduce perspective projection and polynomial unwrap imageing model of central catadioptric cameras with single effective viewpoint, and show calibration algorithm based on polynomial unwrap imaging model. According to the polynomial unwrap imaging model, combining the related features of hyperboloid, propose a simplified polynomial unwrap imaging model which is suitable for omnidirectional vision sensor. The model has less parameters so it can reduce the complexity of calibration.2. Study on the design and epipolar rectification algorithm of binocular ODVS. Design a kind of binocular ODVS which is composed of two ODVS, aligned vertically (coaxial). According to the imaging properties of binocular ODVS of configuration above, propose a kind of epipolar rectification algorithm using vertical lines in structured environments.The algorithm makes the use of constraint of epipolar line better in stereo matching, and reduces calculation complexity of stereo matching.3. Study on the obstacle detection technology using binocular ODVS. A method of obstacle detection is proposed: first process the omnidirectional image with edge detection technology and consider all the edge points detected as the point of suspected obstacles, then process the possible obstacles with stereo matching using constraint of epipolar line, get the height and depth information of the suspected obstacles according to the disparity and geometrical relationship. The possible obstacles are confirmed to be actual obstacles only when the value of height is larger than the threshold.4. Design the prototype of the robot navigation system based on omnidirectional vision. Design a hardware platform of mobile robot and present the software architecture and its implementation. |
PDF Full Text Request