The Research On Mobile Robot Obstacle Detection Based On Vision In Unknown Environment

Obstacle detection, according to the robot's environment information, plays an important role in the research of navigation technology for mobile robot. In this paper, mobile robot's obstacle detection based on vision in unknown environment is researched. It includes three parts: adaptive image segmentation, stereo calibration and obstacle detection based on stereovision.During the part of adaptive image segmentation, the color system is analyzed, and then some kinds of methods for image segmentation are studied. Using the method of adaptive image segmentation, it can detect the suspicious obstacle area primarily, when the color of the road before robot is essentially consistent.Before we use the stereovision method to detect obstacles, we should calibrate the camera system. The camera model and the relation between each coordination system are simply introduced in the part of camera calibration. In the process of experiment, we use the camera system similar to the system of two cameras are parallel, which is get by controlling the movement of the camera and platform. According to this, a camera system calibration method is presented, using the theory of intersection between the two-projection lines. This method is simple and the calculation is sharply reduced.In the part of stereovision, we use stereovision method on the diagonals of the suspicious obstacle areas, which is gained by adaptive image segmentation. And the 3D information can be gained by 3D reconstruction, then the real obstacle area can be extracted from the suspicious area according to the area's height.Through the research of above three parts, we know that if stereovision method is used for every point of the image, the calculation would be complex, but if the adaptive image segmentation is used alone, some false obstacles can't be detected, so a fast feasible obstacle detection method is presented in this paper. It combines the adaptive image segmentation and stereovision method. The algorithm firstly extracts average chroma of reference road in real time and segments the images adaptively to detect the suspicious area primarily. Secondly, it detects the obstacle areas clearly using stereo vision on the diagonals of the suspicious obstacle areas. The corresponding relationship between area's diagonals of right-and-left images is proved when two cameras are parallel. Based on this, the match time can be decreased greatly.Experiments have shown that all these methods are correct and feasible. In this paper, the calculation of camera calibration using Matlab programming, and a mobile robot obstacle detection system is established under VC++ environment. The results show that this system can adapt to the changes of illuminations automatically and quickly detect Obstacles right before the mobile robot.