| With the continuous development of science and technology,intelligent vehicle has increasingly drawn attention widespread.Currently,intelligent vehicle localization solution based on a priori map has been met universal approbation throughout the world.Although the 2D grid map that was more commonly used in the past can directly represent the drivable areas of vehicles,it does not actually describe the surrounding environment commendably.In case of it,2D grid map that is Large and complicated misses a lot of useful environmental feature information.And when it is used for vehicle localization,not only the location accuracy is getting lower,but also more ineffective.Nowadays,the 3D map is mainly used as a priori information to realize vehicle localization.It has more environmental feature information,which can not only be used for high-precision vehicle localization,but also for functions such as environmental perception and path planning.Compared with camera,3D lidar emits laser beams actively to obtain the distance between the lidar and surrounding environmental feature directly.3D lidar has a higher measurement accuracy and a wider measurement range,therefore it is often used to make high-precision 3D point cloud maps.Vehicle localization system based on high-precision 3D point cloud maps does not rely on external satellite signals,which can solve the problem of localization failure well caused by long-term signal loss of integrated localization system of Inertial Measurement Unit(IMU)and Global Navigation Satellite System(GNSS).This paper mainly researches on the algorithm of lidar-based 3D mapping and localization of intelligent vehicle,specifically including lidar data processing and multi-sensor data synchronization algorithm,3D mapping algorithm based on graph optimization and 3D localization algorithm fused IMU and the corresponding experiments.The relevant studies are as follows:(1)Firstly,this paper has mainly studied on lidar data processing and multi-sensor data synchronization algorithm.The principle of lidar is analyzed.Then the point cloud data is pre-processed to remove outliers and points falling on the vehicle body.Next it is down sampled.And then a high-frequency IMU is used to remove the point cloud distortion caused by the rapid movement of the vehicle.Finally,the algorithm realizes the time and space synchronization of multi-sensor data.(2)What's more,the primary coverage of this part is 3D mapping algorithm based on graph optimization and point cloud registration methods.Firstly,the algorithm analyzes and compares the most commonly used point cloud registration methods and selects point cloud registration algorithm based on Normal Distribution Transform(NDT)with better comprehensive performance.Secondly the algorithm proposes a mapping algorithm based on pose graph optimization.By adding the vertices and edges of the key frames,the prior pose edges formed by the GNSS prior data,and the detected closed-loop constraint edges,the entire pose map with constraints is finally optimized to build a 3D map.(3)Moreover,ground segmentation algorithm and 3D localization algorithm fused IMU are the main element of this research.The algorithm uses the concentric circle threshold first to separate most of the non-ground point clouds,and then remove the ground point clouds by fitting multiple planes to obtain non-ground point clouds with less loss.And the algorithm meshes 3D map and stores them,then uses the map manager to load local point cloud maps dynamically.Finally,unscented Kalman filter is used to design a vehicle localization algorithm based on known maps and fusing IMU.(4)An intelligent vehicle experimental platform is built,and multiple sets of point cloud data are collected in the campus scene to verify the above algorithms.The experimental results show that the proposed map construction algorithm can reduce the cumulative error of the map and the fusion localization algorithm can localize intelligent vehicle without GNSS data. |
PDF Full Text Request