Research On Laser SLAM Odometry Based On IMU Tight Coupling And Multi-class Feature Point Matching

Simultaneous Localization And Mapping(SLAM)is one of the key technologies of autonomous driving,providing important guarantee for autonomous driving vehicles to realize environment perception,decision making,and path planning.Due to the accurate ranging of LiDAR and strong anti-interference ability,laser SLAM has gradually attracted the attention of many experts and scholars.However,when the existing method of laser SLAM odometer is used in outdoor dynamic environment,it has the problems of low accuracy and poor robustness due to insufficient matching constraints of point cloud feature points and sparse ground sampling data.To solve the above problems,this paper proposes two odometer methods,including multi-category feature point matching and the tight coupling between IMU and LiDAR.Firstly,in the aspect of feature point matching,the closest point correspondence of each category feature point is searched within the minimum range of Euclidean distance,and the consistency check of the normal vector and principal vector direction of the feature point is carried out to make the constraints in distance and direction more robust.Secondly,in the tightly coupling odometer between IMU and LiDAR,the motion distortion caused by LiDAR data is corrected by IMU state estimation,and the system optimization function is constructed by IMU data to further constrain the pose and suppress the drift in the direction of the gravity vector.The specific research contents and innovations of this paper are as follows:(1)Aiming at the problems of low accuracy and poor robustness of laser odometer caused by insufficient point cloud matching constraints when the existing laser odometer calculation method is used for map construction in outdoor dynamic environment,an odometer calculation method based on multi-category feature matching is proposed.Firstly,in the stage of feature point extraction,dual-threshold ground filtering is used to classify ground points and non-ground points.For non-ground points,multi-category feature points are extracted based on Principal Component Analysis(PCA).Secondly,in the stage of multi-category feature point matching,the closest point correspondence of each category feature point is searched within the minimum range of Euclidean distance,and the consistency check is carried out in the direction the of the normal vector the and principal the vector of feature point,which makes the constraints of feature point matching in distance and direction more robust.Finally,compared with the LeGO-LOAM algorithm,the relative pose error of the proposed algorithm is reduced by 25.5% in the field test.(2)In order to solve the problem of the drift in the direction of the gravity vector caused by sparse ground sampling data and noise in LiDAR data when the existing laser odometer calculation method is used to construct maps in outdoor dynamic environment,an odometer calculation method based on the tightly coupling between IMU and LiDAR is proposed.Firstly,in the stage of data preprocessing,the main target point cloud is extracted,and IMU state estimation is used to correct the LiDAR distortion data to improve the quality of LiDAR data.Secondly,in the stage of feature matching,the consistency check of the normal vector and principal vector of feature points is carried out to make the constraints of feature point matching in distance and direction more robust.Finally,IMU data is used to construct tight coupling and joint optimization estimation to suppress the drift in the direction of the gravity vector,which can reduce the cumulative error of the laser odometer and improve the accuracy of the odometer.In the field test,compared with the LeGO-LOAM algorithm,the cumulative error of the algorithm in this paper is reduced by 63.6%.