The Technology Of Environment Multi-view Stereo Reconstruction Based On Mobile Robot

Based on computer vision in this thesis, and related theory or technology, numerica l simulation was carried out to study the multi- view stereo reconstruction in the environment, and a set of three-dimensional reconstruction model has been developed on mobile robot. The reconstruction process is the ideological basis to SFM, through mathematical modeling of the camera, camera calibration exclude errors and distortion, by matching feature points to find the relationship between the key points, using triangulation and other basics to reconstruct the model, and finally showing the point cloud data by PCL. Entire reconstruction model, the base in C ++ programming language, related to the third-party libraries have OpenCV, PCL and SSBA-3.0, etc., used in the programming platform for Visual Studio 2010, Eclipse and CMake like.Environment multi- view stereo reconstruction mentioned in this paper include three parts which were camera model and calibration, SFM and reconstruction, as well as integrated with the experimental. In the camera model and calibration part, firstly introduced four common coordinate systems, the pinhole camera model with lens distortion, subsequent introduction of rotation matrix and translation vector in calibration, Rodrigues transformation which can transform between the matrix and vector, and the mathematical knowledge Homography. After summarizing the unknown number of parameters, the camera calibration was solved by the Zhang Zhengyo u calibration, and the camera movement position calculated by the trajectory. In the SFM and reconstruction part, firstly introduced the concept of SMF and its application in the common environment, and in stereo imaging, referred three basic concepts which were Triangulat io n, Epipolar Geometry, as well as the Essential and Fundamental matrices. To address the relations hip between the images match, the nature of matching is in order to solve the relationship between the two matrices and feature points. It is necessary to match the feature points, so only with a more classical SIFT algorithm and improved SURF algorithm, and a more rapid Optical Flow method, followed by solving the matrix F and the SVD decomposition. In the case of the previous conditions are obtained, using the basic knowledge, reconstructing the model and then optimizi ng. Obtaining the depth of information from the disparity map, point cloud obtained by calculat io n, and the point cloud database performance with PCL as intuitive 3D models.In the comprehensive and experimental part, summarized the both sides and designed with the corresponding experimental process, access to relevant data by testing, and to be compared with the analysis, such as the differences and the differences between OpenCV and MATLAB. At the end of the experiment are summarized and improved analytical part. The innovation of this thesis is not only multi-view reconstruction of the results obtained model of the target object, but also tracking mobile robot coordinates associated with labeling the trajectories of movement, lack of modeling and various items needed of modeling in the future can be analyzed from a contact in the difference between the two conditions.