Research On Key Technologies Of Unmanned Driving Based On Binocular Vision

With the rapid development of the times and the continuous innovation of technology,people’s requirements for quality of life and environmental are becoming more and more strict.Traditional transportation is gradually abandoned by the times due to its safety and energy consumption characteristics.Instead,it is becoming more intelligent.And Unmanned driving technology came into being in this environment.The key to Unmanned driving technology is its real-time in 3D reconstruction to the scene.However,the actual scene information is complex,and it is difficult to recover the scene information well.Therefore,further research is needed.In this paper,the Unmanned driving technology based on binocular vision is studied,and the stereo matching technology of binocular vision is deeply studied.The main research contents of the thesis are as follows:First of all,the basis theory of binocular vision is researched systematically.And then the theory of the camera imaging and the calibration technology of the camera are researched,and the camera is rectified according to the camera parameters in calibration process.After understanding comprehensively of the calibration algorithm,the calibration technology based on the Matlab calibration toolbox and OpenCV is selected.And complete the encapsulation of a set of system functions,then complete the recurrence of the entire process from calibration to rectify,and the calibration parameters are compared with the parameters of officially provide,and the expected results has been achieved.After that,the ZED operation platform is built.During the construction process,the selection of different versions of the vs platform and the cmake precompilation of the library functions must be considered,so that the internal dll library can be called to prepare for the later 3D reconstruction system.And then the stereo matching technology in binocular vision is researched.Firstly,the basic theory and classification of the stereo matching algorithm are systematically introduced,Including initial matching cost calculation,cost function aggregation,and post-parallax processing,and the evaluation method of the stereo matching algorithm is introduced.Then the semi-global matching(SGM)algorithm is researched,and the BM stereo matching algorithm with CUDA speed-up based on ZED interface is compared.The optimization of the algorithm has been completed.Finally,the secen is reconstructed in real time through the internal 3D reconstruction interface by the disparity map.the recurrence of a set of projects from calibration to 3D reconstruction has been completed.After introducing the semi-global matching(SGM)algorithm,the adaptive support weight(ASW)algorithm in the local stereo matching algorithm is researched.Aiming at the problem that the adaptive support weight stereo matching algorithm(ASW)has poor robustness and high time complexity in low texture and complex texture regions,The Stereo Matching Algorithm Based on Multi-matching Primitive Fusion has been designed,and in the cost function aggregation stage,the aggregation function is designed based on the correlation of the initial matching cost function and pixel value in the reference image.Experimental results showed that compared with ASW,the proposed algorithm greatly improves the accuracy and the real-time performance and higher robustness in complex regions.