Research On Robot Visual Servo Based On Stereo Vision

Binocular stereo vision technology is an important subject in the field of robot visions. The aim of the research is to reconstruct 3D geometry information of the scene, mostly include: image capture, image preprocessing, feature extraction, camera calibration, stereo matching and 3-dimensional reconstruction.Binocular stereo vision's basic theory, method and related techniques have been discussed in this paper, and based on which, MOMOMAN-SV3XL industrial robot visual serving control system is constructed, in which experiments of visual positioning tasks are conducted. The main work in this paper is summarized as follows:1. According to the aim and the device possessed, using two digital camera, M0T0MAN-SV3XL robot and computer to construct the binocular stereo vision system, and achieve the 3D space positioning tasks.2. Using Harris operator to detect the image corner, and provide the pixel coordinates for camera calibration and stereo matching.3. In order to calibrate camera's parameters, based on the Harris operator to detect the image corner, choosing stereo object as reference, using least square algorithm to obtain the projection matrix of left and right cameras. And camera's parameter is achieved by projection matrix. The method to estimate the fundamental matrix is also researched in this paper. Particularly, introducing the normalized eight-point algorithm and making simulation on this method. After estimating the fundamental matrix, constructing the Kruppa equation to obtain the camera's parameter; add the epipolar constraint to stereo matching by fundamental matrix.4. In stereo matching, firstly, using grey cross-correlation coefficients to obtain the matching points set of left and right image, then on account of false match, adding disparity gradient constraint, symmetry testing and epipolar constraint. The result of the experiment shows that the false match could be eliminated on the whole.5. The final purpose of binocular stereo vision is 3D reconstruction. This paper is mainly concerned of the reconstruction of space point. The goal point can be reconstructed by using the matching points of left and right image and projection matrix.6. Adopting object-oriented design and realized technology; selecting C++ language and Visual C++ development environment, this paper develops related software including image capture, target extract, corner detection, camera calibration, 3D reconstruction and robot control system, achieves the MOTOMAN robot stereo locating system.