| In the future,the automobile industry will have four significant attributes,that is,the "four modernizations" that the automobile industry is experiencing at present,that is,networking,electrification,intelligence and sharing.Among them,the automatic driving technology represented by intelligence will experience a long period of development and gradually mature,which will also be the development direction of the automobile industry in the future.The research and application of 3D lidar in the establishment of high-precision map and path planning of intelligent driving vehicle will be the development direction and technical support of automatic driving technology.In addition,due to the stable and powerful performance of lidar,the technical advantages of slam technology in map building and precise positioning are overlapped,at present,real-time location and mapping(SLAM)of 3D lidar is the research hotspot of the core technology in this field and will continue.In this paper,vehicle 3D lidar and vehicle integrated inertial navigation system are used as the main sensors,and the research work of SLAM technology based on 3D laser point cloud is as follows:(1)Analyzing the data structure of the integrated inertial navigation system and lidar,and then preprocessing the original data respectively,which mainly includes the removal of the influence of the gravity acceleration in the data of the integrated inertial navigation system,three-step filtering and the removal of the ground based on the angle differential method for point cloud.Finally,the preprocessed data are fused to remove the motion distortion of lidar.(2)The research and algorithm design of laser radar SLAM are carried out from the front end and the back end.For the front end of SLAM,firstly,the feature points are extracted based on the curvature and RANSAC algorithm,and then the point cloud data is registered by ICP algorithm optimized from iterative solution method.For the back end of slam,the pose graph is built and optimized from two aspects of graph node and node constraint edge,and the loop edge in the node constraint edge is mainly introduced,at the same time,the relocation part is also analyzed.(3)From the aspect of data preprocessing,experiments are designed to verify the algorithm effect of the middle and far end points,outliers,dense points and ground points filtering in the scene,and good preprocessing effect is achieved,which improves the calculation efficiency and robustness of the later data processing;from the aspect of the front end of slam,the point cloud feature point extraction algorithm and point cloud matching algorithm are carried out The experimental verification of accuracy and robustness is carried out.The experimental results show that based on the feature point extraction and matching algorithm in this paper,the final matching accuracy is improved,and the robustness to noise is also strong.From the back-end aspect of slam,the experimental verification of optimization effect and algorithm efficiency is designed.For large scenes,the slam in this paper finally performs well in error control and algorithm In the end,two kinds of open source point cloud datasets,long line scene without loopback and large-scale scene with loopback,are selected to verify the reliability and feasibility of the overall effect of SLAM Algorithm in this paper,and good results are achieved in the details of algorithm processing and overall.Through the analysis of the above experimental results,it is confirmed that the real-time positioning and map building technologies based on lidar discussed in this paper are feasible and reliable in different environments,and achieve the expected results,and can run well on intelligent driving vehicles. |
PDF Full Text Request