Research On Visual SLAM Algorithm Based On Point And Line Features

Visual Simultaneous Localization and Mapping(v SLAM)algorithm for mobile robots as a key technology of mobile robot has always been a research hotspot in the field.Point features are the most widely used features in v SLAM algorithm.But in a low-texture environment,there are fewer point features and more line features.In the v SLAM algorithm,line features can be complementary to point feature information to improve the accuracy and stability of the algorithm.In this paper,we based on the ORB-SLAM2,processing the line features parallel.By fusing the line features,we promised the system is real-time running,at the same time,the system can effectively reduce the tracking failture caused by the reduction of the feature points in the low-texture environment.Firstly,we summarized the development status of visual SLAM algorithm and the development of visual SLAM system based on line features in recent years,and we thought it is valuable for line features to improve the accuracy and robustness of visual SLAM system.we also analyzed the reason why one of the most excellent line feature extraction algorithm in the image is easy to break and proposed an improved line feature matching algorithm,it can effectively reduce the broken line segments calculation of the feature matching in the SLAM system by combing the broken lines.Secondly,we analyzed the problem inaccuracy point and line features cause by noise in the image obtained by the camera,and proposed a algorithm by sampling and fitting line segment,and then comparing the fitted line and the line segment obtained by back projection of the line segment endpoints to filter out line segments which are greatly affected by noise.The algorithm can effectively reduce the problem of inaccuracy line segments’ endpoints.On this basis,we proposed a visual SLAM algorithm of fusing line feature,and verified the proposed algorithm improve the system accuracy and robustness by simulation experiments.Finally,we analyzed the reason why the monocular visual SLAM system is easy to fail in the low-texture environment,and proposed a method of decomposing the homography matrix using line segment endpoints and feature points.By using line feature endpoints to increase the amount of data for decomposing homography matrix,the system initialization robustness had been improved.On this basis,we proposed a monocular visual SLAM algorithm based on point and line feature,the algorithm used epipolar constraints and line feature descriptors to match line features,and verified the improved algorithm improving system robustness and accuracy by simulation experiments.