Research On High-precision Localization Of Vehicles With Tightly Coupled Of LiDAR And Inertial Measurement

Localization is an important part of the automatic driving vehicle technology system.The entire automatic driving system relies on the accuracy and real-time nature of localization technology.Because the Simultaneous Localization and Mapping(SLAM)technology has the characteristics of high accuracy,high frequency,and strong applicability,SLAM is widely used in autonomous vehicle localization.The effects of SLAM algorithms based on different sensors are different.For example,Li DAR is insensitive to light changes and is able to obtain depth information directly.The algorithm that uses Li DAR sensors therefore has better realtime performance,and has stronger resistance to light changes compared to camera solutions.More importantly,the range data of Li DAR is highly accurate,which guarantees the output accuracy of the algorithm in principle.The inertial sensor is not affected by external scenes and can provide stable motion measurement,and the inertial sensor has a high frequency,which can effectively measure information under severe motion conditions.Algorithms using inertial sensors are easier to meet the accuracy and real-time requirements of automated driving systems.Aiming at the Localization scene of autonomous vehicles,this paper proposes a highprecision Li DAR odometry based on feature points matching according to the technical characteristics of Li DAR.Based on the Li DAR odometry,an algorithm tightly coupled with inertial measurement is proposed,which effectively enhances the robustness of the algorithm,and improves the localization accuracy and the quality of mapping.Finally,the proposed algorithm is tested on an opensource dataset and a real vehicle.The specific research content of this paper is as follows:(1)In order to provide accurate extrinsic parameters for the coupled system of Li DAR and inertial measurement,a coarse-to-fine LIDAR-IMU external parameter calibration algorithm with strong adaptive capability is proposed in this paper.First,the hand-eye calibration algorithm is used to initially solve the LIDAR-IMU extrinsic parameters.The algorithm can run stably without relying on the initial values of the sensor extrinsic parameters.Then,the result of coarse calibration of extrinsic parameters is used as input,combined with IMU motion estimation,the point cloud projection error is calculated,the joint optimization function is constructed,and the accurate external parameters are obtained iteratively.(2)Aiming at the problem that the speed of processing large scale point cloud of is slow,and insufficient local representativeness of feature extraction lead by the uneven distribution of the point cloud in the horizontal and vertical directions,a high-precision Li DAR odometry algorithm based on ground segmentation and extraction features is proposed.The algorithm contains a ground segmentation process based on Gaussian process regression.Also,the RANSAC algorithm is used to fit the ground points.And the PCA algorithm is used to estimate the distribution of segmented point cloud.Then,a variety of geometric features with directions are obtained by a feature classification.The nearest neighbor search method is used to obtain the registration error,and the multi-metric method is used to weight the error term,the fixed smoothing method is used in the back end,and the nonlinear least squares are solved by the optimization method to obtain the pose estimation.(3)Aiming at the problems of low frequency of pure Li DAR odometry,the accuracy of the algorithm is obviously reduced under fast and intense movement,and the accumulated error is difficult to eliminate in long distance driving,a tightly coupled algorithm based on Li DAR sensor and inertial measurement is proposed.In this algorithm,IMU motion prediction is used to remove the motion distortion of point cloud,the motion estimation results of Li DAR odometry are taken as a priori,and IMU measurements are treated as motion constraints by using pre-integration method.Then,the tightly coupled motion estimation is obtained by joint optimization.The loop closure detection and GPS measurement based on the descriptor generated by the segmentation method are used to eliminate the cumulative error,and the back-end is managed by the key-frame method based on the incremental smoothing mapping,which update and maintenance of state variables in an efficient way.(4)In order to verify the accuracy,real-time performance and robustness of the algorithm proposed in this paper,the UrbanNav,an opensource dataset,and real vehicle test methods are used to verify the algorithm.The groundtruth in UrbanNav is able to evaluate the location accuracy of the algorithm.In the real word experiment,we use a data collection platform equipped with Li DAR,inertial measurement unit,GPS,industrial computer to collect data at different scenes.The software framework is based on Ubuntu 18.04 and ROS.And algorithm is coded by C++.The algorithm is tested on opensource dataset and self-test data,and the accuracy and robustness of the system are verified to meet the requirements of autonomous driving car.