Research On SLAM Technology Based On Solid-state 3D LiDAR

The mainstream 3D laser sensor in Simultaneous Localization and Mapping(SLAM)technology is mechanical 3D lidar.However,due to its built in rotating parts,the life is relatively short,and the horizontal linear scanning mode is adopted,which will lead to blind areas when identifying the surrounding environment.In contrast,the solid-state 3D lidar has no rotating parts,so it has a long service life.At the same time,the scanning mode without repeated paths can be used to reduce blind spots.Therefore,using solid-state 3D lidar for positioning and composition has gradually become a trend,but the scanning angle of solid-state 3D lidar is limited,and the scanning path is not repeated,which will bring some difficulty to the solution of pose.Aiming at these problems,SLAM technology based on solid state 3D lidar is studied in this paper.The main work is as follows:(1)In view of the data characteristics such as small field of view and non-repetitive scanning path of solid-state 3D lidar,this paper studies the extraction of solid-state 3D lidar point cloud data,and proposes a point cloud threshold segmentation for solid-state 3D lidar filter improvement algorithm.Firstly,point cloud segmentation is used to separate the points in the ground and the field of view.Secondly,the appropriate downsampling filter index is selected according to the different traveling speeds and trajectories of the vehicle.Finally,the comparison experiment before and after preprocessing is carried out to verify the accuracy of the algorithm in this paper to remove noise accuracy.(2)In line and surface feature point extraction,one-time point selection method is usually used to calculate feature points,but this may reduce the accuracy of feature point extraction.Therefore,this paper proposes an improved algorithm for feature extraction and an optimization algorithm for registration.Firstly,the feature points are calculated simultaneously by using two different curvature formulas in the same scene,the selected feature points are combined,and then the nonlinear optimization is carried out to solve the pose and complete the coarse matching of point cloud.Then,the point cloud data is accurately registered by Iterative Closest Point(ICP)according to the vehicle running speed threshold.Finally,through the comparison experiment with the simulated trajectory,it is verified that the algorithm in this paper has a better registration degree.(3)With the accumulation of time,the cumulative error of solid-state 3D lidar becomes larger.When the solid-state 3D lidar recognizes a scene that has been visited before,there will be a position offset because the scene cannot be recognized.Therefore,this paper proposes an improved closed-loop detection algorithm based on graph theory to optimize the map.The selection of key frame point cloud data is optimized by setting the time difference,and the speed and distance co-view threshold method is set to perform ICP registration to optimize the closed-loop detection pose.By comparing the mapping effects before and after closed-loop detection optimization in the same scene,the importance of closed-loop Detection optimization for mapping accuracy is verified.Based on solid-state 3D lidar,this paper improves the traditional algorithm in data screening,feature point extraction,point cloud registration and closed-loop detection,which is beneficial to improve the performance and practicability of solid-state 3D lidar.