Research On Mapping And Localization Method Of Underground Garage Based On The Fusion Of LiDAR And IMU

In recent years,with the further development of new energy vehicle and artificial intelligence,autonomous driving has gradually entered real life.In the urban underground garage scene,the realization of home zone parking or valet parking requires accurate vehicle localization information.The traditional vehicle localization mainly depends on the Integrated Navigation System(INS),but in the underground closed scene,The lack of Global Navigation Satellite System(GNSS)signal and the cumulative positioning error of Inertial Measurement Unit(IMU)driving for a long time lead to poor stability and accuracy of vehicle localization results.Relying on the localization method of a priori map,especially the point cloud map with high precision and stability,can well solve the problem of accurate vehicle localization in closed scenes.Based on the above analysis,this thesis mainly studies the point cloud map construction algorithm based on the fusion of Li DAR and IMU,and the vehicle localization algorithm based on a priori point cloud map.The specific research contents are as follows:(1)The data preprocessing is completed for Li DAR and IMU.The data characteristics of lidar and the mathematical model of IMU are analyzed,and the time synchronization between Li DAR and IMU is completed based on this.The online extrinsic calibration of Li DAR and IMU is realized based on the principle of hand eye calibration.With the help of time synchronization and calibrated IMU data,the motion distortion of Li DAR point cloud data is compensated.(2)The mathematical model of SLAM,the principle of pose graph optimization and incremental smoothing are studied.The construction of laser odometry for underground garage is realized.Aiming at the lack of loop detection in LOAM framework,a loop detection algorithm based on Scan Context and historical frame strategy is designed to improve the mapping accuracy.The IMU pre-integration constraint is designed to improve the stability of mapping.Finally,a tightly coupled SLAM framework of Li DAR and IMU based on graph optimization is constructed.(3)Aiming at the initial localization problem of a priori map,a global pose initialization method based on feature description and random sample consensus is designed.Based on the derivation of vehicle constant turn rate and velocity motion model,the vehicle localization algorithm integrating point cloud prior map and IMU is realized by using Extended Kalman Filter algorithm.(4)The above algorithm is verified by experiments on the built real vehicle platform.The experimental results show that the motion compensation based on IMU data has higher accuracy.Compared with the front-end slam method,the slam framework built in this thesis has better stability and accuracy in map construction.The localization method based on a priori point cloud map can effectively solve the problem of vehicle localization in the scene of underground garage,and has higher localization accuracy,and can meet the needs of automatic parking.