Research On 3D Reconstruction Based On Model Registration And Binocular Stereo Vision

Three-dimensional reconstruction based on image is a computer technology using image to restore three-dimensional information of objects, including image acquisition, image preprocessing,3D reconstruction, model registration and other steps. Binocular vision has become a research hotspot in the computer vision domain because of it's simple and reliable device and accurate reconstruction. Based on the theory of computer vision, the thesis focuses on the research of camera calibration, stereo matching,3D reconstruction of binocular vision system and model registration. The main research work and achievements are as follows:In this thesis, stereo vision model was firstly studied. The binocular stereo vision model can be divided into general binocular stereo vision model and horizontal parallel binocular stereo vision model. As the latter form of construction is as same as the human being's eyes system, so it is the most widely used binocular stereo vision model currently. So, this thesis designed the binocular stereo vision system for objects reconstruction and the reconstruction steps were designed. This thesis was focused on the solution process of the Zhang Zhengyou planar template calibration method, and calibrated the binocular stereo vision system.Secondly, in this thesis, stereo matching algorithm was studied in-depth. The stereo matching is based on certain rule to find the correspondence between pixels in different images from the same space point. It is the most important and difficult problem, and the matching accuracy is directly related to the results of three-dimensional reconstruction. The stereo matching algorithms can be divided into local and global stereo matching algorithm based on the optimization theory. In this thesis, the representative stereo matching algorithms were studied. The former included the classical correlation window and adaptive weight-related window algorithms, and the latter included the stereo matching based on graph cuts algorithm. After the comparison by experiments, we concluded that the precision of stereo matching algorithm based on graph cuts was higher and it was suitable for accurate reconstruction of objects.Thirdly, the theory of three-dimensional reconstruction and model registration were studied in this thesis. We can restore three-dimensional model of objects after the disparity map and the interior and exterior parameters of cameras were obtained. Rather than obtaining all the three-dimensional information of object only in one place, we need to obtain the image from multiple views and to conduct a three-dimensional reconstruction from each view. In order to obtain overall three-dimensional model data of the object, we need to put all the pieces in the same coordinate system using rigid transformation. As a result, we can obtain a overall model of object. This process is called the model registration. In this thesis, the RANSAC and ICP model registration algorithm was studied in depth, and the overall model of object was obtained.Finally, the work was summarized and the prospect was proposed in this thesis.