Development Of High-precision Localization And Environment Construction Algorithm For Automatic Driving In The Park

Self-driving technology is undoubtedly one of the hottest technologies today,as driverless cars become a reality on the road,Internet companies such as Baidu have tried in this field one after another.Technology enterprises are the first to propose to carry out L4 level unmanned driving with fixed routes in a closed environment,and driverless vehicles in certain parks have become the fastest way for driverless vehicles to land.The key to solve the autonomous driving in the park is high-precision map and high-precision localization.Therefore,this project takes the park as an automatic driving development scene,and in view of the problems of unstable positioning information and positioning drift in the current high-precision localization,such as weak GNSS signal,3D lidar is used as the main sensor,and inertial sensor is integrated to develop the high-precision localization and environment map construction algorithm.Based on the project of the ministry of science and technology of the People's Republic of China,"key technology research and demonstration operation of electric autonomous vehicle",a high precision localization and point cloud map generation algorithm is proposed to meet the requirements of automatic driving in the park.Specific research contents are as follows:(1)In order to solve the problems existing in 3D lidar,such as data redundancy,high computational pressure,the inability of sparse point cloud to accurately represent environmental information,and the processing of unstructured point cloud information such as ground,occupying computing resources,etc.In this study,in order to reduce the computational pressure of subsequent algorithms and facilitate inter-frame matching and feature extraction of point cloud,a point cloud data preprocessing algorithm for synchronous positioning and mapping technology is proposed: the lidar data frequency is selected to match the autopilot scene in the park,the data is sampled by voxel filter,the threshold segmentation of the region of interest in the point cloud and the extraction and classification of the point cloud by Euclidean distance clustering are carried out.From the data comparison effect,it can be seen that the preprocessing greatly reduces the amount of data processing calculation of the subsequent algorithm,reduces the computational pressure from the practical application level,and lays an effective foundation for the subsequent research environment construction and the high-precision localization algorithm.(2)In view of the problem of environmental reconstruction in the park,considering the real-time requirements of synchronous localization and mapping algorithm for data processing,this paper comprehensively analyzes the current mainstream ideas of point cloud feature extraction,and proposes a point cloud feature extraction method based on curvature segmentation which is effective and realtime.Furthermore,point cloud scan-scan matching is realized for feature points.Compared with the standard point cloud registration method that iterates the nearest point,the computation amount of this algorithm is reduced,and the robustness of the algorithm is improved.Local point cloud map is constructed by point cloud registration.Aiming at match point cloud registration error,which is likely to cause "ghosting" of the map and subsequent positioning drift,map optimization framework is adopted to optimize the global map under the support of local map.According to the results of ICP iteration,the edges and vertices in the graph optimization are constructed,and the constraint equation is solved by g2 o optimization to generate the global environment map in the park scene.(3)Aiming at the problem of high-precision positioning in the park,simply using point cloud map matching for positioning has the cumulative error of positioning,which affects the positioning accuracy of long-distance driving for a long time.Therefore,this study uses the data high-frequency characteristics of the inertial measurement unit to pre-integrate the motion information before the arrival of the next frame point cloud to estimate the vehicle's current pose.At the same time,the constraint equation of feature line and feature plane is established from the feature point of point cloud,which is transformed into the optimization solution of nonlinear least square problem.Finally,the joint positioning optimization of inertial measurement unit and lidar is used to update the vehicle posture.In addition,the optimization results are used to update the state variables in the inertial measurement unit prediction and eliminate the drift noise of the inertial sensor.(4)Finally,this study uses the laboratory intelligent drive-by-wire vehicle hardware platform and Robot Operating System(ROS)software platform to design test scenarios to realize the algorithm real car test and algorithm improvement optimization.The ROS is used to realize the interaction between the computing platform and the sensor data,the information transmission between each child node,and the positioning information in the form of CAN signal to the vehicle decision level.Experimental results show that the positioning and mapping algorithm can not only provide accurate positioning information,but also effectively make up for the shortage of GNSS positioning method under the weak GNSS signal,and to a certain extent,the cumulative error can be eliminated by comparing with the current mainstream lidar synchronous positioning and mapping method.Under the typical working conditions of Nanling campus of Jilin University,the environmental point cloud map can be constructed accurately to eliminate the "ghosting" of the map and meet the requirements of map accuracy.Based on the analysis of the application results of autonomous driving,the research content of this paper can be applied to the L4 level autonomous driving in the park,to achieve trajectory tracking and more unmanned driving tasks,and to ensure the safe and smooth operation of unmanned cars in the park.