Research On A Low-Cost MEMS IMU-Assisted Lidar SLAM Algorithm

Driverless vehicle is one of the most popular research directions in the vehicle field in recent years.High-precision positioning and high-precision map construction are the key to realize driverless vehicle.The simultaneous localization and mapping(SLAM)technology based on Lidar can realize high-precision positioning and high-precision map construction.However,in the complex and changeable road environment,Lidar SLAM algorithm still has many shortcomings in the accuracy of feature extraction,the accuracy and real-time of positioning and mapping.In order to improve the performance of Lidar SLAM algorithm,this paper studies the advanced LOAM(Lidar Odometry and Mapping)algorithm framework,focuses on the feature extraction,location and mapping in the algorithm framework,adds loop detection and back-end optimization to the algorithm framework,and further studies the lidar SLAM algorithm assisted by MEMS(Micro-Electro-Mechanical System)IMU.The specific research contents are as follows:(1)According to the research on the feature extraction module in the LOAM algorithm framework,it is found that the feature extraction algorithm using average curvature sorting has the problems of discrete point interference and low feature extraction accuracy.By analyzing the characteristics of mechanical lidar point cloud,it is proposed to segment the point cloud according to the scan line according to the characteristics of the lidar point cloud,and then segment and classify the scan line point cloud into interior points,edge points and discrete points.Extracting linear feature points and plane feature points from the scanning of objects in the point cloud can effectively avoid the interference of discrete points and improve the accuracy of feature extraction.(2)Aiming at the research on the positioning and mapping module in the LOAM algorithm framework,it is found that the use of grid maps to store point cloud maps has the problems of slow map update rate and poor real-time frame-to-map matching.Through the analysis of the storage and establishment principle of point cloud map,it is proposed to store the point cloud map by storing key frame pose points and their corresponding key frame point clouds.The frame point cloud is projected to the map coordinate system to establish a point cloud map.Finally,the frame-to-map matching method in the LOAM algorithm is used to estimate the pose,which can effectively improve the map update rate of the point cloud map and improve the real-time performance of frame-to-map matching.(3)In view of the problem of serious cumulative error caused by no loopback detection and back-end optimization in the LOAM algorithm framework,this paper adds loopback detection and back-end optimization to the positioning and mapping module of the LOAM algorithm framework.The loop is determined by the distance and time difference with the nearest key frame pose point,and then the back-end optimization method of GTSAM graph optimization is used to correct the deviation of the historical key frame pose.(4)For the research of lidar devices,it is found that the lidar has point cloud distortion and lidar has the problem of failure scenarios.By analyzing the characteristics of lidar and MEMS gyroscope devices,it is proposed to use low-cost MEMS gyroscope to output high-frequency angular velocity and acceleration information to correct lidar point cloud distortion,and to use MEMS gyroscope to assist lidar SLAM algorithm to run in failure scenarios.The proposed SLAM algorithm is experimentally verified by the independently collected campus environment data set and KITTI data set in various environments.The experimental results show that the proposed SLAM algorithm based on low-cost MEMS IMU assisted Lidar effectively improves the accuracy,robustness and real-time performance.