Research On Lidar Localization Technology Assisted By Semantic Map

Robot autonomous positioning and navigation is an important research direction in the field of robots.The robot positioning method can be divided into local positioning such as getting the relative position from SLAM odometer based on the Markov chain and getting the absolute position from the global positioning if there is an additional assisted map.The local positioning method will cause the accumulation of positioning errors due to its positioning principle;and the map contains the precise position information of the targets in the environment,so this method can provide a reliable absolute position.In the current forms of map,grid maps,topological maps and other forms can only give geometric information of the environment.However,in addition to geometric information,semantic maps also provide higher-level semantic information of the environment,that is,the category and location information of environmental targets,which can help robot understand the environment and achieve a high level of human-computer interaction.If there is a semantic map containing semantic information of the global environmental static targets in advance,then extract the semantic information of the environmental static targets scanned by LiDAR(Light Detection and Ranging)in the current position,and match it with the global semantic map,we can get the absolute localization of LiDAR.This semantic map-assisted localization method can also give the robot a reliable reference localization,ensuring that the robot can continuously locate and perform tasks delivered by humans.This thesis is based on the idea which uses semantic map to assist LiDAR localization.The main work done is as follows:(1)Use the LiDAR point cloud data processing algorithms to extract the semantic information of the static targets in the environment,that is,the trees.Use the point cloud filtering algorithms to remove outliers in the original point cloud,compress massive point cloud data,etc.;remove the ground point cloud from the filtered point cloud data to prevent the ground point cloud from interfering with the extraction results of static target information;use the Euclidean Cluster algorithm and Sampling Consistency segmentation algorithm to extract the semantic information of the static targets of the environment.(2)For the semantic information of the environmental targets extracted by the semantic extraction algorithms,use the angle threshold to filter out the target clusters when the angle between the cluster and the ground is too small,use the distance threshold to filter out the target clusters that do not meet the distance requirements,then,the semantic reference point is obtained.These reference points have reliable position information for matching and positioning.(3)Pre-construct a global semantic map containing the semantic reference point information of static targets in the environment,and apply the semantic extraction algorithms described above to generate semantic data frame for the point cloud scanned at the current position of LiDAR.Then match the semantic data frame with the global semantic map to achieve the purpose of LiDAR positioning.(4)Use the outdoor environment data collected by LiDAR equipment to perform positioning experiments,and evaluate the accuracy and efficiency of LiDAR localization method assisted by the semantic map proposed in this thesis for the positioning results of some calculation examples.The positioning results show that the algorithm time-consuming of the semantic map-assisted localization method proposed in this thesis has greatly improved,up to 46%,and this method can provide a stable and reliable global reference localization in terms of localization accuracy.Experiments show that the semantic map-assisted LiDAR localization method proposed in this thesis is feasible and can provide a robust localization result.