Research On Lidar SLAM Method With Fused Point Cloud Intensity Information

In recent years,with the rapid development of the autonomous driving industry and the introduction of new concepts such as smart cities,autonomous driving systems have received more and more attention from the public.The perception system is the "eyes" of autonomous driving,and is one of the most important basic technologies for autonomous driving systems.The development of SLAM technology is of great value in promoting the application of various autonomous driving platforms on the ground.Among perception systems,laser SLAM is widely used because of its high accuracy,ability to work around the clock,and scene adaptability.Laser SLAM,a technology that uses laser sensors for real-time environment building and mobile system localization,is widely used in robot navigation,UAVs,autonomous driving,virtual reality,and other fields.The current mainstream laser SLAM algorithms only use the geometric information of the environmental scene for point cloud alignment,pose estimation and loopback detection,ignoring the characteristics of the unique intensity information of Li DAR point clouds,and when the amount of point cloud data is large and the autonomous mobile platform is moving fast,the number of point cloud frames increases rapidly,and the accumulation of point cloud alignment errors is fast,which easily leads to serious distortion of3 D point clouds;meanwhile,the existing loopback detection algorithms have a large drift in the At the same time,the existing loopback detection algorithm is prone to the failure of loopback detection due to the large drift of the moving platform trajectory results.To address the problems of laser SLAM,the research in this paper is as follows.(1)A point cloud alignment algorithm that fuses point cloud intensity information is proposed.In this paper,the intensity information of the point cloud and the geometric information of the environmental scene are fused to construct a globally consistent environmental feature descriptor,and a non-iterative two-step method is used to find the nearest neighbor of the point cloud in the point cloud alignment stage.Firstly,we use the geometric information for coarse finding and coarse alignment,and then use the point cloud intensity information for secondary finding and fine alignment,so as to improve the robustness and accuracy of the point cloud alignment.(2)A global feature descriptor-based loopback detection method is proposed.By fusing the point cloud intensity information and using the already constructed globally consistent descriptors,the laser point cloud descriptors are extracted by using circular partitioning in a dimensionality reduction way,which ensures the rotational invariance of the environment descriptors.And combined with the domain-based search method,the search of the loopback frames is completed,and finally the Intensity-ICP is used to complete the fine alignment of the loopback frames and output the optimal positional transformation,so as to complete the loopback detection.(3)Application validation of the improved laser SLAM algorithm for mobile platform.The experimental platform used in this paper is a self-developed autonomous mobile platform,and the improved laser SLAM algorithm is validated using public data sets,while the point cloud data of the campus environment is collected and tested,and the robustness and generalization of the improved laser SLAM algorithm is validated.