Research On Odometry And Mapping Method Of Lidar Inertial Coupled

State estimation is the key technology of mobile robot positioning and navigation.Through multi-sensor fusion,it can make up for the lack of single sensor for state estimation,provide more reliable position and attitude information,and provide important guarantee for real-time control,decision-making and path planning of the robot.SLAM technology can realize the self motion estimation of mobile robot in the unknown environment,that is,"positioning",and establish the map of the surrounding environment at the same time,which has become a research hotspot.In this paper,for the outdoor dynamic road environment without GNSS signal,aiming at the problems of low accuracy and insufficient robustness of existing laser odometry methods in this challenging scene,a odometry and mapping scheme of 3D Lidar and MEMS inertial navigation coupling is proposed.In order to ensure the real-time performance of the system,data preprocessing,feature extraction of dynamic obstacles,laser odometry and laser mapping are performed on different threads.Compared with the most popular LOAM and Lego-LOAM schemes,this paper makes the following improvements:Firstly,in the process of feature extraction of dynamic obstacles,the dynamic obstacles in the road environment,such as pedestrians and vehicles,are removed by using the depth image method.For each frame of point cloud data,the ground is removed,the remaining point cloud is segmented and clustered,and the dynamic obstacles are removed after geometric judgment.Then,the static point cloud of each frame is extracted from the ground feature points,nonground line feature points and non-ground surface feature points,and classified for storage.Second,in the Lidar odometry,the continuous frame matching is carried out,and the extracted three types of feature point clouds are correlated according to certain order and classification,that is,the ground feature,non-ground line feature and non-ground surface feature point clouds of the current frame are registered with the corresponding ground feature,non-ground line feature and non-ground surface feature point clouds of the previous frame respectively,and are matched between frames After that,we get the rough position information(three translation and three rotation).Then,the Error State Kalman Filter frame is used to fuse the pre integration information of imu and the matching information of Lidar frame,further improve the accuracy of pose estimation,and estimate the zero deviation error of accelerometer and gyroscope in imu,and the corrected imu can compensate the nonlinear motion distortion of the next frame of laser point cloud.Finally,using the data collected from real dynamic road environment and the dataset of laser odometer published by KITTI,qualitative and quantitative accuracy analysis of the proposed scheme and loam and Lego-LOAM schemes are carried out.The running time of each module in the three schemes is evaluated with the measured data.The test results show that the accuracy of Lidar odometry is better than that of LOAM,which is equivalent to Lego-LOAM.