Research On LiDAR High-precision Positioning Algorithm For Vehicles

With the continuous popularization of autonomous driving technology and the continuous improvement of people’s living standards,autonomous vehicles have gradually become popular.Autonomous driving is a complex system that integrates multiple functions such as precise positioning,environmental perception,dynamic decision-making and planning,motion control and execution.Precise positioning is the basis for these functions.LiDAR adopts laser ranging technology and accurately measures the distance of obstacles around the carrier.It has the advantages of high measurement accuracy,wide detection range,high time and spatial resolution,and rapid response to environmental changes.It can support accurate positioning and environmental perception,which is one of the most important sensors for autonomous vehicles.LiDAR can autonomously perform relative positioning,or achieve absolute positioning with the support of a priori high-precision map,and it can also easily perform information fusion with vehicle-mounted GNSS/INS/cameras.The research of precision positioning algorithm gradually realizes the lane-level positioning ability in the urban environment,which is the basis for the fusion of multi-source sensors for autonomous driving.Our main research work is as follows:1)This paper first introduces a LiDAR SLAM algorithm based on edge and surface features,and designs the front-end optimization strategies,which can ensure the calculation efficiency of the system and improve the accuracy of both positioning and mapping.On this basis,this paper introduces a multi-thread optimized NDT matching algorithm,designs the strategy of raster storage and dynamic loading of map,and improves the efficiency and accuracy of map matching process.2)In order to improve the absolute accuracy of the point cloud map and reduce the point cloud dislocation and ghosting in the map,this paper combines the advantages of LiDAR SLAM and POS mobile surveying system,designs a point cloud map generation scheme based on pose graph optimization,and realizes the construction of high-precision point cloud map with good consistency in the global coordinate system.3)In order to reduce the interference of dynamic obstacles to point cloud mapping,this paper proposes a dynamic object removing method.According to the relative pose relationship between the previous and current key frames,the static point cloud is marked in the coincidence area of two frames,and only the static points are stored in the high-precision map.The measured data show that the stability and reliability of the whole positioning system can be improved after the dynamic objects are removed.4)Aiming at the problems that relative positioning suffers from error accumulation and absolute positioning is seriously disturbed by the environment and sensitive to the initial pose,this paper proposes a map aided feature ICP + NDT positioning algorithm based on the advantages of both relative positioning and absolute positioning.KITTI data is used to verify that the statistical accuracy of the proposed method is about 10 cm,which meets the requirements of lane-level positioning.5)In order to verify the effectiveness of the proposed positioning algorithm in this paper,a hardware platform of LiDAR/GNSS/INS and other multi-sensor integration is built.The measured data show that the overall accuracy of the algorithm in the two kinds of designed scenarios is roughly the same,the absolute error of horizontal positioning is less than 30 cm,and the absolute error of vertical positioning is less than 5 cm;In the closed park scenario,the overall relative positioning accuracy is about 1.18%,and the heading angle attitude accuracy is about 0.45 deg;In the open road scenario,the relative positioning accuracy of our method is about 1.24%,and the heading angle attitude accuracy is about 0.65 deg.Decimeter-level positioning of a single LiDAR in an urban environment is achieved.