Research On Measurement Technology Based On Binocular Stereo Vision

Binocular stereo vision is among the crucial branches of machine vision research.With the enhanced processing power of computer hardware and accuracy of image acquisition equipment,it has been widely used in industrial,aerospace,medical,and automotive fields.Binocular vision measurement technology can obtain the threedimensional spatial information of the shooting target without contact,and has the advantages of simple structure,convenient operation and accurate measurement.The paper studies the method of binocular stereo vision measurement from stereo vision technology,and builds a binocular vision measurement system by modifying the stereo matching algorithm in order to upgrade the accuracy of the measurement.At the present stage of medical image acquisition using CT equipment,physicians are requested to ascertain the patient’s area of the inspection and the scanning scope on the basis of clinical scanning parameters,manually position the patient then release the line to gain medical images.In this research paper,the length and thickness of the human examination area are measured in a non-contact condition to facilitate the remote movement of the scanning bed.The theoretical and experimental analysis is presented for the vital techniques involved in the measurement process of binocular vision measurement system,including binocular camera calibration,image pre-processing,stereo correction,as well as stereo matching.The primary research of this paper are as follows:Firstly,the geometrical model of binocular camera imaging,the theory of binocular vision measurement as well as the principle of binocular camera calibration are illustrated.The camera calibration experiments of Zhengyou Zhang’s calibration method are executed,which finally selects the automatic calibration approach based on OpenCV library programming to acquisition of camera internal and external parameters.The reprojection error of that approach is relatively minor.Secondly,the Contrast Limited Adaptive Histogram Equalization image enhancement algorithm and the Bouguet correction algorithm are selected to deal with the images gathered by the binocular camera.In this way,to eliminate the problem of the difference between brightness and darkness of the image as well as the inability to align the pixel points of the left and right images due to the structure of the camera its self,uneven lighting and lens distortion.Then,the crucial techniques of the stereo matching algorithm are studied,and the SGBM algorithm is selected as the stereo matching algorithm for this study after experimental comparison and analysis of the matching effects of the three commonly used stereo matching algorithms,BM,SGBM and AD-Census.For the problems of uneven image light and darkness,poor stereo matching quality at the left and right imaging blind areas,and isolated voids in the acquired parallax map,the SGBM algorithm is improved by combining the gradient cost and Census cost as the matching cost method to calculate the final generation value,and refining the parallax map by adding the Adaptive Median Filtering algorithm in the parallax post-processing step,which improves the capability of the algorithm to resist light interference and the accuracy of matching.Finally,the binocular vision measurement experiment platform is built,and the MFC graphical interface is designed based on Visual Studio 2019 and OpenCV vision library to complete the dimensional measurement experiment in the CT machine debugging room.The experimental outcomes show that the measurement system which is built in this study is capable of accurate measurement,and the relative error of measurement does not exceed 0.62%,which can provide accurate data for the movement of scanning table.