Research On Technology Of 3D Information Acquisition Based On Binocular Stereo Vision

Stereo vision is one of the most important distance perceptive technologies in the field of computer passive ranging. After it was born, with more than 30 years development, stereo vision has been widely used in robotic vision, aerial mapping, military application, medical diagnosis and industrial detection. Binocular stereo vision is one of the important branches of stereo vision. It simulates scene processing of human vision directly, so it can measure the 3D information of scenes flexibly in various conditions. Research of Binocular stereo vision has great significance for vision physiology and engineering application.This paper focuses on key techniques of binocular stereo vision, such as camera calibration, image rectification, stereo matching, 3D reconstruction, and main works are accomplished as following:Firstly, the present situation and research methods of binocular stereo vision in our country and abroad are elaborated and summarized, advantages and defaults of each method are also analyzed, and all this make an important basis for the research in the paper. Secondly, pin-hole camera model, perspective transform calibration and stereo image rectification based on perspective projection matrix are introduced in detail, then camera calibration experiment is carried out, and internal and external parameters of binocular camera are obtained.Thirdly, local stereo matching algorithm and acceleration techniques are researched, matching tactics in actual application are explored, effect of the window size of local stereo matching algorithm on matching results is analyzed, the application of dynamic programming and belief propagation in stereo matching are studied, and stereo matching algorithm based on belief propagation is improved, experiment results show that the algorithm improved can converge within less iteration with lower mismatching rate.Finally, space point reconstruction method with different camera configuration is introduced, and then an indoor scene is reconstructed by utilizing two parallel cameras.