Research And Implementation Of Outdoor SLAM Based On LiDAR And Inertial Navigation Couplin

State estimation is a prerequisite for autonomous platforms such as autonomous vehicles to realize the functions of navigation,obstacle avoidance,and planning.Simultaneous Localization and Mapping(SLAM)technology based on multi-sensor fusion can achieve higher precision vehicle autonomous localization and incremental map construction in unknown environments,so it has been widely concerned in recent years.After analyzing the characteristics of outdoor dynamic road scenes,this thesis proposes a 3D LiDAR and inertial navigation system(INS)tightly coupled SLAM algorithm to solve the problems of low accuracy and insufficient robustness of current LiDAR Inertial Odometry in this scene.Using multi-threaded real-time completion of point cloud feature extraction based on dynamic and static point segmentation,LiDAR odometer,LiDAR and inertial navigation tightly coupled graph optimization,loop closure detection,and other steps,compared with the currently popular algorithms to make the following improvements:(1)This thesis proposes a real-time dynamic and static point cloud segmentation method for LiDAR and INS coupled SLAM systems.The dynamic and static points of the point cloud are segmented by using the distance image projected by the point cloud,the point cloud residual image of adjacent frames generated based on Inertial Measurement Unit pre-integration and the depth network.Experiments show that point cloud features are only extracted from static objects,and the movable but static objects in the scene are retained to reduce the inter-frame feature correlation error caused by dynamic objects.(2)The scan-to-map matching strategy is improved,and a point cloud matching method combining feature distance and reflection intensity is proposed.The optimization part of the global factor graph of the backend of the LIO-SAM algorithm is changed to a sliding window optimization strategy based on keyframes.Through tight coupling optimization,IMU and LiDAR are combined to construct the motion equation and observation equation for state estimation,to achieve higher precision pose estimation compared with the original algorithm.Experiments on self-collected data and public data show that compared to the original algorithm,the output pose error can be effectively reduced(3)This thesis proposes a more robust loop closure detection algorithm and based on the Scan Context loop closure detection algorithm,the principal component analysis is used to establish a correction local coordinate system to reduce the mismatch caused by the change of viewing angle.This thesis improves the generating strategy of feature descriptors,adds the distance information to judge the similarity of the point cloud,and proposes a new similarity calculating formula.The improved loop closure detection algorithm is integrated into the SLAM scheme proposed in this thesis.The experimental results show that the improved loop closure detection algorithm achieves better results than the previous one.