Lidar Localization In Garages For Intelligent Connected Vehicles

With the development of the Intelligent Connected Vehicle(ICV)industry,more and more humanized service requirements about driverless vehicles are put forward and needed to be achieved.Among them,the most critical localization solves the fundamental problem of "where is the car" and is closely related to most vehicle-mounted applications.In practical application scenarios,not only the real-time and high-frequency localization results are required,but also many localization results are required as data input for many vehicle-mounted applications,so as to make humanized choices for relevant information.Location-based assist in ensuring the safety of remote driving,push more accurate information related to media services,and exchange location information between different vehicles or roadside units.In the outdoor,ICVs can obtain precise localization through Global Navigation Satellite System(GNSS).However,this method cannot be applied to scenes where the vehicle is driving indoors,such as in the garage.At this time,the GNSS satellite signal is blocked and the precise location cannot be obtained.On the other hand,due to the dim light of the garage,the similarity between the wall and the column is high,resulting in the problem of low texture without obvious features in the garage.3D lidar is not affected by the indoor environment,and is especially suitable for garage scenes with low-texture garages.Therefore,we design a localization system only based on the lidar in the indoor garage.Processing and optimizing the point cloud data collected by our vehicles in the garage scene can effectively solve problems such as the localization error.Guarantee high frequency and high precision localization results in the low-texture garage,in order to provide ICV corresponding localization.The system improves the Simultaneous Localization And Mapping(SLAM)method,and proposes corresponding solutions to the three typical problems of low-texture,long-tailed and repetitive structure in the garage,which have been implemented.(1)Based on prior map,the final localization is obtained by combining the frontend local pose estimation result of high-frequency intensity optimization with the global matching result of high precision key frame and prior map back-end.(2)In the local pose estimation,for the low-texture problem,the intensity information of lidar point cloud is introduced to carry out relative static object filtering and obtain meaningful feature points.Aiming at the low-texture problem,feature extraction based on field of view segmentation is carried out.Local pose is estimated by an intensity optimized odometery.(3)In the global matching,aiming at the problem of repeated structure,global localization is carried out based on prior map.coordinate system is unified through the determination of initial pose when the system is called for the first time.The current frame is globally matched with the local map based on the estimated pose,and mismatching is avoided through global optimization.The main advantage of this system is to extract the point cloud feature points of meaningful,distinctive features and high-reliability areas,making the matching results more accurate.According to the collected data and the actual vehicle experiments,compared with the existing localization system,our system can effectively improve the robustness of localization,improve the accuracy of localization results,and reduce the cumulative drift,error,and delay.Finally,centimeter-level positioning results of realtime output with the same frequency as radar input are obtained.