Research And Development Of Binocular Vision 3D Reconstruction System For Autopilot

With the rapid development of computer vision,artificial intelligence and deep learning have also been widely developed.There are more and more researches on the direction of environmental perception such as vision.The research on 3D reconstruction of binocular vision has also received extensive attention in various fields.Binocular vision is to perceive the surrounding environment by simulating the structure of human eyes,and to transform the perceived two-dimensional image information into three-dimensional information of space through stereo matching.Binocular vision can be widely used in unmanned environment perception,security environment detection,AR 3D reconstruction and other aspects.However,due to the existence of illumination,occlusion,weak texture and other conditions in the environment,the actual matching effect of the stereo matching algorithm is not ideal,so the 3D reconstruction based on binocular vision still has very important research significance.In this paper,binocular vision is used as the environment perception sensor in the scene of low-speed unmanned mobile platform to complete the three-dimensional reconstruction of the surrounding environment of the unmanned mobile platform.In this paper,the binocular calibration,stereo matching and 3D reconstruction of point cloud data involved in the 3D reconstruction of binocular vision are studied.The error correction of the current calibration method is carried out,the binocular stereo matching algorithm is improved,the parallax map is obtained,and then the 3D reconstruction is carried out.This paper first systematically introduces the theoretical basis of binocular vision,and then calibrates the checkerboard lattice binocular by two methods,to obtain the internal and external parameters and distortion coefficient of binocular camera.The parallax map is obtained by using improved cost function in stereo matching.Finally,the 3D reconstruction results are obtained by using the parallax diagram.The main work of this paper is as follows:(1)select the hardware sensors that meet the requirements of binocular vision on the unmanned mobile platform.The hardware platform of binocular vision 3D reconstruction system of unmanned mobile platform is built.(2)analyze the camera calibration principle and camera distortion,obtain the internal and external parameters and distortion coefficient of the camera through stereo calibration using MATLAB binocular calibration toolbox,and conduct error analysis and correction on the results obtained.(3)the stereo matching algorithm was studied in depth.AD-Census was used as the cost function,and the calculation method of bit string in the cost function was improved.Then use the cost to calculate the cost and the variable window cost aggregation,use the multipath optimization parallax diagram.Thus,a relatively stable parallax diagram can be obtained.(4)Designed the software framework of the system,and combined the commonly used libraries and development tools in visual processing to realize the call to the hardware interface,the reading and saving of images,the realization of stereo matching algorithm,and the realization of 3D reconstruction.