3-D Evaluation Techniques For Rock Surface Based On Stereo Vision

One of the most important mission in planetary exploration is to sample and to analyze rock and soil from other planets surface by using rovers. The grinding work for hard material such as rock is usually be done by vehicle manipulator so as to remove the external covering layer, and use relevant equipment to sample and to analyze. The key technology in the mission is to reconstruct the 3-D information of rock surface based on binocular vision and to search for the approximate flat surface of rock automatically by use of relevant algorithms so that the 3-D locating point coordinate and locating normal vector could be obtained in order to guide vehicle manipulator to locate the rock surface accurately and to grind it. This paper sample the image of rock through binocular stereo vision, and use Matlab calibration toolbox to compensate for the distortion of camera so that the camera projection model could approach to ideal pinhole model.Besides that,the paper also compare the method with the traditional DLT calibration method and the Tsai two step calibration method. In order to improve matching speed, the image rectification is also necessary to be done. The phenomenon of inhomogeneous stretching will appear who could influence matching accuracy in rectified image after projection transformation. Therefore, the paper adopt stereo matching method based on the inverse transformation for epipolar line rectification. The method open the search window in the rectified image, and then inverse rectify the image coordinate for the points within the window so that the gray value of pixel in un-rectified image is extracted and the gray correlation operation is computed between reference point and matching point, this method decreased mismatching rate effectively caused by interpolated point. In addition, the paper adopted two-way matching, which take the first matching point as a new reference point and search for the new matching point in reference image, if the new matching point and the original reference point are same point then we consider that the matching is correct at this time. This paper also adopt a fast searching strategy of matching for the sake of improving matching speed according to principle that the projection points of space point located on different position. Firstly, the ground data and the background data are separated by clustering distance after obtaining space coordinate of scene. Then, the cloud point data of rock are triangulated by the method of Delaunay triangulation, and calculated normal vector of each triangular plane. Finally, the angle between normal vectors are clustered and analyzed so as to evaluate the flat surface of rock.