Research On Target Pose Measurement Technology Based On Large Field Of View Binocular Vision System

With the rapid development of computer vision technology,binocular vision measurement technology is widely used in industrial and military fields such as autonomous driving and video surveillance.The field of view of common cameras at this stage can no longer meet the requirements of most scenes,so fisheye cameras with a large field of view are more and more favored.Taking marine surveillance as an example,the fisheye lens has a short focal length,a large field of view,and a wide field of view,which can realize monitoring with a large field of view.However,the imaging results of the fisheye lens are seriously distorted,and the target matching is difficult.At the same time,due to environmental factors,the base of the binocular vision system cannot be guaranteed to be fixed,which leads to the relative movement of the camera coordinate system and the geodetic coordinate system.The target position coordinates of are no longer the real coordinates in the geodetic coordinate system.In view of the above problems,this paper studies the target pose measurement technology of the large field of view binocular vision system under the moving base.Firstly,the positioning measurement principle and tilt compensation method of the binocular vision system under the moving base are analyzed.The commonly used stereo vision coordinate systems and their transformation relations are studied,and the target positioning formulas in binocular parallel mode and binocular convergence mode are deduced.Aiming at the situation that the base swings,a tilt compensation method is proposed,which restores the coordinates in the swaying camera coordinate system to the initial coordinate system,and builds the target positioning model under the moving base.Secondly,the calibration method,distortion model and correction algorithm of the binocular fisheye camera are studied.In view of the situation that there are too many invalid areas in the fisheye calibration image and the corner extraction speed is slow,the original calibration process is improved.The effective area of the calibration image is extracted first,and then the corner detection is performed,and the corner coordinates in the effective area are restored to The camera calibration is performed after the original position,which reduces the corner extraction time and speeds up the camera calibration.Two kinds of distortion correction algorithms,latitude and longitude correction and camera calibration correction,and the binocular stereo correction method are studied,and the experimental verification is carried out and the experimental results are given.Then,the stereo matching technology of binocular vision system is studied.The basic principle and classification of stereo matching are introduced,and the SGBM algorithm and ORB feature matching method with high real-time performance are analyzed.Aiming at the problem that it takes too long to eliminate the false matching points in the ORB feature matching process,an improved registration method combining GMS and RANSAC algorithm is designed by using the bidirectional matching strategy.Comparing the above algorithms,the feature matching algorithm is selected to complete the stereo positioning,and the advantages of the improved matching algorithm in matching speed are verified.Finally,a large field of view binocular system is built for experiments.The accuracy and speed of binocular positioning are verified when the base is fixed.When the base is rocking,the pose is measured to verify the correctness and effect of the tilt conversion compensation method proposed in this paper.