Research On Lidar-based Outdoor Large-scale Location Method

With the rapid development of the unmanned vehicle industry,how to realize autonomous positioning of unmanned vehicles in a wide range of outdoor environments has become a current research hotspot.In a large-scale environment,as the moving distance increases,the cumulative error in positioning obtained using sensors such as lidar and Inertial measurement unit(IMU)will gradually increase,and more sensor information is needed to achieve accurate positioning of unmanned vehicles.When there are few features in the environment,the registration algorithm based on feature matching will lack enough feature points for registration between point clouds.This article mainly focuses on the autonomous positioning of unmanned vehicles in a large-scale outdoor environment.The method of first establishing an offline point cloud map and then using the point cloud map to achieve unmanned vehicle positioning is to use Global Positioning System(GPS),IMU,and 3D lidar as sensors.The focus is on the construction of offline point cloud maps.The positioning based on offline point cloud maps and the correction of map errors without GPS are used.The positioning method in this paper effectively realizes the precise positioning of unmanned vehicles,which has important research significance.The main research work of this paper is as follows:(1)Aiming at the outdoor large-scale environment with few features,a mapping method based on the Open MP-boosted Normal Distributions Transform(NDT-OMP)registration algorithm was designed.The algorithm uses GPS,IMU,and 3D lidar as sensors,and uses GPS information to optimize the registration process of the NDT-OMP registration algorithm,and uses the General Graph Optimization(g2o)framework to add constraints on GPS information and IMU information to the registration algorithm.The results are optimized to complete the construction of the offline map.(2)In the absence of GPS,a positioning algorithm based on Unscented Kalman Filter(UKF)is designed.This algorithm uses the offline point cloud map as the reference frame for registration,combines the UKF algorithm and the NDT-OMP algorithm,uses the predicted value obtained by the prediction process of the UKF algorithm as the initial registration value,and registers the current point cloud with the offline point cloud map.The obtained registration result is used as the observation value of the UKF algorithm.After the observation of the UKF algorithm is updated,the final pose estimation is obtained.(3)Aiming at the process of mapping process,this paper proposes a map error correction algorithm.The algorithm first records the landmarks at locations with many environmental features.During the positioning process,the closest landmarks are found based on the current distance between the point cloud position and the landmarks.The Iterative Closest Point(ICP)algorithm is used to achieve the registration between the current point clouds and the landmarks.The registration result meets the conditions,and the current point cloud and road landmark are calculated to obtain the map correction amount.Finally,the pose obtained by the positioning algorithm is modified to obtain a modified pose estimate.