3D Reconstruction Based On Binocular Stereo Vision

Recently, artificial intelligence, intelligent driving, robotics, unmanned aerial vehicles,aerial photography, virtual reality, augmented reality and other fields have strode into a new research boom. The study based on binocular vision is an important branch in the field of machine vision, which plays an important role in promoting the rapid development of intelligent assisted driving, robot path planning and navigation, UAV and other fields. Therefore, the research of binocular 3D perception reconstruction technology has important practical significance. Based on the Marr vision development theory, the binocular 3D reconstruction technology is studied in this paper. The main research contents are as follows.(1) The binocular vision system is studied. Based on the parallel vision model, the selection of camera, the selection of webcam and the checkerboard pattern are studied, and finally a binocular vision experimental platform is built.(2) High quality images are obtained by image preprocessing. Several classical denoising algorithms are introduced to denoise the image, and then the image is sharpened and enhanced to reduce the smooth and fuzzy influence caused by denoising.(3) Feature point detection is an important part of machine vision research. This paper introduces the common and effective feature point detection algorithm, and analyzes the advantages and disadvantages of each algorithm. On this basis, this paper proposes a multi-scale Harris corner extraction algorithm.(4) This paper introduces the conversion relationship between four coordinate systems,analyzes the commonly used models, introduces several types of calibration algorithms, and analyzes their respective applications and advantages and disadvantages. Based on the study of nonlinear models and Zhang's calibration algorithm, a three-step calibration method based on second order mixed model is proposed.(5) In the 3D reconstruction system of binocular vision, the most difficult step is stereo matching, the core idea of which is to seek the corresponding relation between the projection points of the same scenic spot in different images. Firstly, this paper analyzes the knowledge of matching primitives, constraint criterion and matching similarity degree, and then introduces the commonly used matching method, analyze the advantages and disadvantages of various algorithms. Based on this, a cost variable S-census algorithm is proposed.(6) The basic principle and the implementation process of 3D reconstruction are introduced.At last, this paper makes a good reconstruction of the actual scene through experiments on the established platform of binocular vision experiment system.