Research And Developmention Of Binocular Stereo Video Acquisition System

Binocular stereo vision has been widely used in many fields,such as unmanned aerial vehicle obstacle avoidance,robot navigation,precision industrial measurement,object identification,virtual reality,scene reconstruction,exploration and so on.This thesis takes the recorded binocular two-dimensional video as the research object,and carries out the research of transforming binocular two-dimensional video into three-dimensional point cloud video.The main research work is as follows:(1)Zhang's calibration method is analyzed and improved.Based on Zhang's calibration method,this paper calibrates the camera parameters,and uses black and white chessboard targets to calibrate the camera.in the experimental process,it is found that the pixel error of the first calibration is relatively large,which makes it impossible to obtain accurate camera internal and external parameters.Therefore,on the basis of the original method,taking the distortion coefficient obtained from the first calibration as a reference,and surrounding the distortion coefficient,the second calibration is carried out.The experimental results prove that the internal and external parameters of the camera obtained after the second calibration are lower than those of the first calibration in terms of pixel error,and the related internal and external parameters of the camera can be obtained more effectively.In addition,according to the internal and external parameters of the camera obtained after calibration,polar line correction is carried out on the images taken by the camera,so that two images of left and right viewing angles reach the result of pixel line alignment,which relieves the calculation pressure for subsequent matching work.(2)The semi-global matching algorithm is mainly studied and improved.Based on the semi-global matching algorithm,this paper uses SSE4.2 instruction set to improve the performance of the original matching algorithm.The original matching algorithm is transformed from serialized calculation to parallel calculation,so as to adapt to the performance problems caused by the gradual improvement of image resolution.At the same time,combined with experiments,it is proved that compared with the original method,the improved method has no influence on the matching effect of the original matching algorithm,but it further improves the performance of the original matching algorithm.In addition,median filtering processing is also carried out on the original disparity map obtained after matching,which further solves the problem of black spots.(3)The conversion process of converting 2d video into 3d point cloud video is studied.In this paper,the obtained disparity map is converted into a three-dimensional point cloud model,and at the same time,orb feature point extraction algorithm is used to extract feature points from two frames of images,and one of them is converted into a three-dimensional point cloud model.according to the relative positional relationship of feature points in two frames of images,the rotation matrix and displacement matrix of the two frames of images converted into the three-dimensional point cloud are calculated,and according to the rotation matrix and displacement matrix,two three-dimensional point cloud models are spliced,frame-by-frame converted and frame spliced,so as to achieve the final synthesis of the three-dimensional point cloud video.(4)A binocular stereo video acquisition system is designed and developed.This system realizes the function of converting two-dimensional video collected in binocular devices into three-dimensional point cloud video.