Research On SLAM Algorithm Fusion Of Visual Point Line Features And Lidar Point Clouds

Mobile robots have been widely used in many areas of human life,such as automated warehousing,cleaning services,medical care and other tasks.Among them,Simultaneous Localization and Mapping(SLAM)plays an important role in mobile robots,which enables robots to navigate and localize autonomously in unknown environments.Lidar SLAM and Visual SLAM are two common SLAM methods.Lidar SLAM has high precision and strong robustness,and can work under low light conditions;its disadvantage is that it is difficult to extract semantic information from lidar point clouds and the price of lidar is high.In contrast,Visual SLAM is cheap and easy to use,but it is easily affected by ambient lighting and field of view,and it is difficult to deal with scenes with insufficient lighting and lack of texture.In order to overcome the shortcomings of the SLAM algorithm based on a single sensor and improve the robustness and accuracy of the algorithm,this thesis studies the SLAM algorithm that integrates visual Point-Line features and lidar point cloud.The main research content and completed work are as follows:(1)Propose lidar and visual fusion SLAM framework.Aiming at the shortcomings of the Lidar SLAM algorithm and the Visual SLAM algorithm,a lidar and vision fusion SLAM framework is proposed and a lidar and vision fusion strategy is designed.The fusion strategy mainly includes: Visual SLAM initialization strategy based on lidar odometry,visual feature reconstruction strategy based on lidar point cloud,distortion correction strategy of lidar point cloud based on visual odometry,and loop detection strategy based on visual aid.The experimental results show that the fusion strategy proposed in this thesis can effectively improve the performance of the SLAM algorithm.(2)Propose an improved line segment feature extraction algorithm.Aiming at the disadvantages of the original line segment detection algorithm,such as slow extraction speed and easy extraction of unstable line segment features,a hidden parameter adjustment strategy and a line segment length threshold limit strategy are proposed.Aiming at the geometrically adjacent broken lines in the image,a broken line merging algorithm is proposed.The experimental results show that the improved algorithm can effectively obtain stable line segment features and the processing time of a single frame image is 2.4 times faster than the original LSD algorithm.(3)Fusion of visual line features.In order to improve the robustness and accuracy of the algorithm,spatial line reprojection constraints are introduced in the back-end optimization of Visual SLAM.The main design is the spatial linear reprojection residual,and the Jacobian matrix of the spatial linear reprojection residual is derived.(4)Algorithm performance test based on M2 DGR dataset and self-built dataset.On the M2 DGR dataset,the positioning accuracy and robustness of the algorithm proposed in this thesis are superior to SLAM algorithms such as Vins-Mono,PL-VINS,and LIO-SAM.The average root mean square error of the algorithm in this thesis is 0.998 meters,which is about20%,80%,and 86% lower than the average root mean square error of algorithms such as LIOSAM,Vins-Mono,and PL-VINS.On the self-built dataset,the maximum translation error of the algorithm proposed in this thesis is about 0.674 meters.