Integrated Localization Method Of Automatic Driving In Closed Park Based On LiDAR/IMU Tight Coupling

As a common means of transportation,automobiles bring great convenience to people’s daily production and life.At the same time,they also have many problems such as poor safety,low efficiency,and serious fuel consumption.The research of autonomous driving technology has become a hot spot and trend in the automotive industry today,especially the research on realizing the autonomous movement of vehicles and completing some established tasks in a relatively complex outdoor environment,which can make it play a greater role in transportation and other fields to achieve higher value,and can effectively alleviate the above problems.In recent years,the development of autonomous driving technology in closed parks is on the rise.Relatively closed areas such as ports and factory parks have become hot spots to realize the commercialization of autonomous driving technology because of their fixed driving routes and single environment.Autonomous vehicles usually need to have environmental perception,autonomous positioning,planning and decision-making,as well as vehicle control capabilities.Among them,the capability of autonomous positioning is an important basis for autonomous vehicles to achieve autonomous movement in complex environments.At present,most automatic driving systems use positioning methods based on global navigation satellite systems/inertial navigation systems(GNSS/INS).GNSS/INS positioning relies on the radio signals of long-distance orbiting satellites,the positioning accuracy in the closed park of the weak GNSS signal cannot be guaranteed.And INS relies on inertial measurement unit(IMU)for positioning,which has the inherent defect that the positioning error diverges with time.In recent years,the localization method based on simultaneous localization and mapping(SLAM)has gradually become a research hotspot because of its low environmental sensitivity and other advantages.The slam technology of light detecting and ranging(LiDAR)is less affected by illumination and has the characteristics of high robustness in highly complex environment.It helps to solve the problem of error divergence caused by unstable signal strength when autonomous vehicles use GNSS/INS for positioning in closed park.However,the existing SLAM technology based only on LiDAR has problems such as inaccurate initial values,unstable feature point extraction,large vertical errors,and easy to fall into local optimum.The positioning accuracy and robustness in closed environment cannot meet the requirements of practical applications.To this end,this subject adopts a multi-sensor fusion positioning scheme,takes the LiDAR/IMU tightly coupled autopilot combined positioning method in the closed park as the research object,and uses the LiDAR feature point extraction method based on variable neighborhood and feature classification to extract stable,highly representative and efficient feature points to fully ensure the positioning accuracy of the laser odometer.The IMU data is processed based on the pre-integration strategy to ensure the accuracy of the estimated pose using the IMU data,while improving the real-time performance of the system.The LiDAR and IMU data are fused based on the tight coupling strategy,and use the vehicle kinematics model to constrain the vehicle pose to improve the accuracy and robustness of the automatic driving positioning system in the closed park scene.First,the laser odometer is designed in two parts: feature extraction and point cloud registration.The feature extraction part uses the variable neighborhood strategy to determine the feature point extraction neighborhood of the corrected point cloud,extracts a certain number of feature points based on the feature extraction threshold,and classifies the proposed feature points according to the curvature to make it represent different types of characteristics.The point cloud registration part searches for the corresponding points of the selected feature points through the KD tree,and based on the distance to indicate the relative relationship between the feature points and the corresponding points to construct the posture transformation residuals of the autonomous vehicles between adjacent times,so as to obtain a more reliable High laser odometer.Then build the IMU pose estimation model.Pre-integrate the IMU data with the LiDAR measurement time as the time reference,and use the rotation constraint method to estimate the IMU random walk.At the same time,the vehicle pose based on the IMU data is constructed on the basis of fully considering the error transfer relationship between the IMU noise and the pre-integrated quantity estimate the residuals.Next,build a prediction model of vehicle motion.Based on the traditional two-dimensional vehicle kinematics model,the system state quantity is used to construct the model input quantity,and the three-dimensional vehicle motion attitude prediction model is constructed through plane geometry,Ackerman steering principle and kinematic equations,and the prediction residuals are obtained on this basis.In this way,the six-degree-of-freedom pose prediction of the autonomous vehicle between adjacent LiDAR measurements can be realized.In order to avoid the possible negative impact of time offset on positioning accuracy and reliability,the multi-sensor time synchronization method based o n compensation optimization is further discussed,the time offset of LiDAR and IMU is compensated as additional state variables and be jointly optimized with other state variables of the system,using GNSS timing at the same time,and synchronizing the data time of different sensors based on the running time of the algorithm to obtain an accurate fusion time estimate.At the same time,based on the precise synchronization of the time of each sensor,the multi-module data of the laser odometer,IMU pose estimation model,and vehicle motion pose prediction model data are fused on the basis of the tight coupling method with the time reference of LiDAR measurement,and the space reference of IMU coordinate system,and the residuals of each module are jointly non-linearly optimized,so as to obtain high-precision and robust positioning results in the closed park scene.Finally,the actual vehicle test and result analysis based on the data of the closed park scene are carried out.The autonomous driving test platform vehicle collects multiple sets of sensor data in closed park scenes,compares and analyzes three different positioning methods from three aspects according to the test results to fully verifying the effectiveness of the positioning method in this paper.The test results show that compared with the current mainstream pure laser positioning method A-LOAM and the LiDAR/IMU tightly coupled positioning method without adding vehicle kinematics models,the LiDAR/IMU tightly coupled positioning method with additional constraints of vehicle kinematics model can achieve robust positioning results with higher accuracy and stability.