Research On Mobile Robot Localization Algorithm Based On Binocular Vision

As one of the emerging products of artificial intelligence,intelligent robot has a very broad future market.How to achieve robot positioning is one of the core problems of robot intelligence.For now,Simultaneous Localization and Mapping(SLAM)is currently the most mainstream robot positioning algorithm.Visual SLAM means that the input information of the system is only the visual information(image or video stream)from the camera,and the camera posture and other information are calculated through the visual information.Visual SLAM based on the feature point method is a relatively common robot positioning method.In this method,even if the camera movement is too large,as long as the matching point is still in the pixel,it will not cause no matching phenomenon.However,in the matching process of feature points,the phenomenon of mismatching is common,which reduces the performance of the system.Moreover,in some special scenes,such as blurry images,lack of textures,or changes in lighting caused by camera rotation or moving too fast,the visual positioning system based on the feature point method is not satisfactory.In view of the above problems,this paper studies the mobile robot vision SLAM algorithm.The principle and method of ORB-SLAM2 visual positioning algorithm based on point features are studied,including front-end visual odometry,back-end nonlinear optimization,closed-loop detection and map construction.First,the principle and method of camera calibration are analyzed.Secondly,based on image feature extraction,the ORB point feature extraction method and principle are studied,and some feature matching methods are introduced.Aiming at the problem of mismatch in the feature matching process,an improved KNN-RANSAC feature extraction and matching algorithm is proposed.Experiments show that the algorithm improves the accuracy of feature matching and improves the system's accuracy on the basis of little impact on real-time Overall performance.Aiming at the problem that the traditional ORB-SLAM2 system is prone to camera blur when the camera rotates or moves too fast,resulting in blurry images and missing textures,a visual positioning algorithm based on point and line features is studied.Using LSD algorithm for line feature extraction,and for the excessive segmentation of LSD line feature extraction,an improved line segment feature merging algorithm is proposed.Experiments show that this algorithm reduces the time-consuming process of line feature matching and improves the real-time performance of the algorithm.In addition,the RANSAC algorithm is used toreduce the mismatch in the line feature matching process,and high-quality line feature matching results are obtained to estimate the camera pose,thereby improving the overall performance of the system.Finally,an experimental simulation platform was built using a personal computer and tested using the KITTI data set test to verify the superiority of the improved visual positioning algorithm based on point-line features.