Localization Method And Application For Intelligent Vehicles In Congested Urban Environment

Vehicle localization is a key technology for autonomous navigation in urban environment.Urban road environment has rich structural characteristics,and LIDAR has excellent environmental modeling ability,which is very suitable for intelligent vehicle localization in urban environment.However,in the congested urban environment,there are a large number of vehicles and people on the road,these objects are highly dynamic,and will produce serious occlusion.In this kind of congested city road,LIDAR based localization method is affected by dynamic jamming and serious occlusion,which will lead to the difficulty of map building and map matching.In addition,navigation in complex urban environment requires a good real-time localization algorithm.The traditional localization method based on point cloud often suffers from a large amount of map data and poor real-time performance.Therefore,the robustness and real-time performance of the localization results are the two major difficulties in the application of LIDAR localization method in urban congested environment.By analyzing the characteristics of point cloud data at different heights,this thesis proposes two intelligent vehicle localization methods based on octree and cylindrical features respectively,to improve the robustness and real-time of LIDAR positioning in congested cities.For the trees,buildings and other objects around the road,these objects are generally higher than vehicles,and are not affected by the dynamic occlusion.The octree based localization method selects the aerial point cloud for mapping and positioning,which ensures the robustness of the localization algorithm.Aiming at the real-time problem of point cloud matching,a fast ICP matching method based on octree is proposed in this thesis.Through the establishment of octree structure to achieve an effective spatial index,improving the search efficiency of the corresponding points in the ICP algorithm.And by using the hierarchical characteristic of octree and the multi-resolution matching strategy,the matching accuracy is guaranteed and the computation is furtherly reduced.The localization experiment in this thesis proves that the proposed method has good robustness and real-time performance in complex urban roads.The point cloud in the middle of the road environment is affected by dynamic vehicle occlusion,so it is necessary to extract stable and reliable features for localization.In this thesis,the column feature which is robust to vehicle occlusion is selected as the localization feature,and an intelligent vehicle localization method based on columnar feature is proposed.The method extracts columnar features such as trunk,utility pole from middle part of the point cloud for mapping and positioning.In order to extract the columnar feature from the point cloud quickly and robustly,two columnar feature extraction methods based on 3D grid and annular scanning are proposed,and the observation error model of columnar feature is established by circle fitting.Finally,a particle filter algorithm based on columnar features is proposed to realize the real-time robust localization of intelligent vehicles in urban congested environment.The experiment of localization in congested city environment proves that the proposed method based on columnar feature ensures the real-time,robustness and accuracy of the localization results,and greatly reduces the data volume of map.