Research And Implementation Of Depth Perception Technology Based On Binocular Vision

Computer vision is one of the most popular directions in the field of artificial intelligence.As one of the key technologies of computer vision,binocular stereo vision makes the computer perceive the real world directly by simulating the human visual system with low cost,simple equipment and flexible use.It is an important means to obtain the depth information of 3D scene.It is widely used in various fields of production and life,such as robot navigation,Remote sensing measurement,driverless car,aerial survey and game industry.However,due to the limit of some factors and conditions,binocular vision technology can not really be realized as common as human vision,and its accuracy of depth perception is directly affected by the accuracy of stereo matching.This paper will focus on the depth information acquisition,matching algorithm of binocular stereo vision system and so on.Aiming at the depth perception technology based on binocular stereo vision,an experimental platform is set up and the experiment is simulated.The experimental results verified the accuracy of depth information extraction,the main work is as following.Firstly,the binocular stereo vision system model and some basic theoretical knowledge of stereo matching has been introduced,the general process of stereo matching and the standard of stereo matching evaluation are described.Secondly,in the aspect of local stereo matching algorithm,the adaptive support weighting algorithm is studied,and the classical adaptive support weighting algorithm is improved.Taking into account the strength of grouping by similarity is set up on CIELab space,the CIELab color space gray absolute difference and Census transform are used as a measure function of the adaptive weighting algorithm.The experimental results show that the improved algorithm proposed increases the matching accuracy and solves the block effect produced by the traditional algorithm in the low texture area.Thirdly,in the aspect of global algorithm research,a dynamic programming algorithm based on multi-path vertical constraint is proposed in this thesis.The algorithm proposed is used to solve the problem of streak flaw occurring in the disparity map obtained by the original dynamic programming stereo matching algorithm and the low matching accuracy at the edge of the image.In the dynamic programming path optimization stage,the vertical constraint is introduced,and the multi-path backtracking strategy is used to store the disparity path under the threshold to guide the energy optimization of the next row.The experimental results show that the proposed algorithm effectively eliminates the drawbacks of traditional algorithms and improves the overall matching accuracy.At last,the experimental process based on binocular stereo vision depth perception is designed and a hardware platform is set up.The algorithm proposed in this thesis generates the disparity map of scene.According to the triangulation principle,the depth information of the scene images and the point cloud data are obtained.The scene is reconstructed and displayed.The results show that the depth information extraction has high accuracy and the reconstruction effect is excellent.