Research And Implementation Of Key Technology In Binocular Stereo Vision Measurement

In recent years,with the continuous breakthrough of image processing and computer technology in theory and practice,there're wide fields using computer vision technology,including robot obstacle avoidance,unmanned,non-contact measurement and space exploration,so computer vision technology has become a research hot spot which domestic and foreign universities and scientific research institutes attach importance to.The main sensor of computer vision is camera,it's cheaper compare with other sensors.So the study of computer vision technology not only the theoretical value,but also have a broad industrial and commercial prospects.Binocular stereo vision is an extremely important branch of computer vision,which based on parallax theory to achieve 3D measurement and positioning,through the left and right camera simulation of human eyes.The main research content of measurement and positioning based on binocular stereo vision,including camera calibration,binocular stereo correction,stereo matching and 3D reconstruction.Therefore,this paper begins with binocular stereoscopic vision hardware system,mainly studying the binocular calibration and stereo correction,summaring and analysing the existing matching algorithm,further researching and improving this algorithm,and ultimately achieving high-precision measurement.In the paper,with respect to the hardware system of binocular stereo vision,focusing on the problem of target accuracy and resolvability,proposing a theoretical model of binocular inherent error based on geometric relation.Analyzing the relation between hardware parameters and inherentness of the system.Thus,completing of the hardware parameters of the design.Because of low precision,strict camera moving requirement and so on of traditional calibration and self-calibration in stereo vision calibration.In the paper,we mainly study the calibration algorithm of Zhengyou Zhang,introduce the principle and implementation process of the algorithm in detais.And completion of the binocular calibration based on monocular calibration,analyzing the reprojection errors.We research Bouguet correction algorithm to realize the stereo correction.About the problems of disparity discontinuities,weak texture and occlusion in stereo matching.There're are many solutions,but it's difficult to take into account the accuracy and real-time.In the paper,we redesign the similarity measure function and introduce the image filtering operator based on semi-global matching algorithm.The effectiveness of the improved algorithm is verified by the image provided by the Middlebury test platform.In this paper,we implement the complete binocular stereo vision measurement system,using the system for measurement in actual scene.The experiment result shows the high accuracy of the measurement in a certain shorter range.