Research And Development Of Mobile Robot Stereo Vision 3d Reconstruction System

Based on the theories of stereovision and projective geometry, the technology of 3D reconstruction is studied, and a 3D reconstruction system of the stereovision of the mobile robot has been developed in this thesis. This system studies a certain scene carefully in which the mobile robot is interested. The left and right cameras capture the images of the same scene from different views by the movement of the stereovision platform. The images are used to calibrate the parameters of the camera, match the features and calculate the depth and 3D coordinates of interesting points.The 3D reconstruction system includes three components: camera calibration, feature correspondence and 3D information calculation. In camera calibration, a perspective model of the camera is set up and the camera parameters, such as the foundation matrix, the projection matrix and the homography matrix etc., are introduced. In Zhang's calibration method, a known template is needed as a premise which is unpractical to the mobile robot. Therefore, a new method of camera calibration based on the characters of anti-symmetric matrix is put forward in the thesis. The new method does not restrict the movement of camera severely and needs no template and can meet the requirements of mobile robot. In feature correspondence, Harris Corner is adopted to extract features at first. By maximal correlation coefficient of features of left and right images, the rough matching corners are found at second. However, there are many false pairs. At the end, in order to resolving this problem the epipolar geometry of stereovision described as fundamental matrix is used to eliminate the false pairs and find new pairs at the same time. After the camera models and the projection of spacial points in two images are gained, the depth of the point can easily be calculated by using the triangle theory. Given an error in projection equation, the 3D information of spacial points can be calculated by least square algorithm. When the object that creates the error changes, better results can be obtained by linear iterative method.In this thesis, a new camera calibration method based on character ofIn this paper, a new camera calibration method based on character of anti-symmetric Matrix, a features' matching method based on Harris Corner and the epipolar geometry, and a linear iterative method to get 3D information are advanced respectively to three parts of 3D reconstruction system. These methods are all proved correct and feasible by experiments. As a result, a software prototype for 3D reconstruction is implemented by combination of Matlab and C++, and OpenCV(Open Source of Computer Vision) Lib. which is developed for application and development of computer vision by Intel Inc.. The prototype realizes the purpose that the scene where mobile robot roams is reconstructed, and provides a foundation for more far-ranging functions of stereo vision on mobile robot.