Point Cloud Matching Positioning Based On Vehicle Lidar

Positioning is one of the key issues of autonomous navigation technology for autonomous vehicles.Maintaining stable,accurate and real-time positioning can ensure the safety of autonomous vehicles.The main positioning methods for autonomous vehicles include: positioning methods based on the Global Navigation Satellite System(GNSS),positioning methods based on lidar point cloud map matching,etc.At present,in the field of autonomous driving,the global satellite navigation system is the most widely used and most mature positioning technology.Autonomous vehicles can complete precise positioning and navigation in most scenarios rely on GNSS positioning technology.However,there are still situations where the satellite signal is blocked by obstacles such as buildings and trees in the surrounding environment,resulting in positioning failure or offset.At present,in order to improve the positioning accuracy,the Real-time Kinematic(RTK)in GNSS combined with the inertial navigation system(INS)is used for multi-sensor fusion positioning to meet the needs of high-precision positioning of autonomous vehicles.However,the interference to positioning in a complex terrain environment is still inevitable.Therefore,the positioning method based on the global satellite navigation system in the field of automatic driving is still greatly restricted in the scene of complex terrain environment.The point cloud maps collected by the on-board lidar can not only provide a full range of road feature information,but also provide a data basis for centrifugal positioning of autonomous vehicles.The positioning technology based on point cloud map matching can be independent of weather,terrain and other environmental factors.It can be independent of each other in positioning calculations.It can replace GNSS positioning in complex environments and provide guarantee for the safe driving of autonomous vehicles.To this end,this paper studies the positioning technology based on point cloud map matching,and improves the classic point cloud matching algorithm,compares the positioning accuracy and efficiency of different algorithms in complex environments,and verifies the feasibility of positioning technology based on point cloud map matching.The research data of this paper is selected from the urban environment point cloud data in the largest international autonomous driving scene data set(KITTI).First,pre-process the point cloud data,use the LOAM algorithm to construct a point cloud map in advance,as the basic data of the matching positioning experiment,use the filtering algorithm in the PCL library to filter and optimize the noise points in the point cloud data,and enhance the matching Stability of positioning;Then,the iterative closest point(ICP)algorithm and the normal distribution transform(NDT)algorithm are used to perform a single frame point cloud and point cloud map matching positioning experiment.Analyze the stability,accuracy and efficiency of the two algorithms;Finally,based on the current completion of matching positioning,research and algorithm improvement are performed,the Gauss-Newton method is used to improve the ICP algorithm,and the KD tree is used to accelerate the search for the corresponding point,and integrated with the NDT algorithm.It is defined as the NDT-ICP algorithm to perform matching positioning experiments.The experimental results show that the improved NDT-ICP algorithm combines the high precision of the ICP algorithm and the high efficiency of the NDT algorithm.The average distance between the corresponding point between the point cloud to be registered and the source point cloud map is 4.52 cm in terms of positioning accuracy.It can meet the positioning needs of autonomous driving vehicles.In terms of positioning efficiency,the average time spent by each single-frame point cloud for matching positioning is 6.47 s,and there is room for improvement in the real-time positioning of autonomous driving vehicles.