Research On The Theory And Method Of SLAM Integrating GNSS/IMU/LiDAR/Vision

Compared with the traditional vehicle-borne mobile mapping system,the handheld SLAM(Simulataneous Localization and Mapping)system can move freely at any time during the onsite data collection process,and the data can be obtained wherever the personnel can pass,which has low requirements for the working environment and strong adaptability.And it can well make up for the blind zone of satellite remote sensing technology and aerial remote sensing technology.But for the handheld SLAM system,it still has severe challenges of traditional SLAM.The complex practical application environments with rapidly changing illumination,fast motion,scene structure degradation and textureless,etc.,can cause performance degradation of the SLAM system.The handheld SLAM system itself has special requirements on the computational intensity,accuracy,consistency of state estimation,and limitation period of the algorithm.Different sensors can provide global or local positioning results and rich perceptual information for the handheld SLAM system.This thesis investigates how to improve the robustness,sensing capabilitiesand accuracy of the localization and mapping algorithm by fusing multiple sources of information,while maintaining its real-time performance.The main contents and contributions are as follows:(1)A multi-GNSS(Global Navigation Satellite System)satellite clock linear regression prediction model with short historical data is proposed,and the evaluation of multi-GNSS satellite clock extrapolation error and its impact on real-time PPP(Precise Point Positioning)positioning after communication interruption is highlighted.Aiming at the problem that the satellite clock product is missing due to communication delay or interruption during the GNSS real-time PPP operation,performing real-time PPP initialization again increases the convergence time and affects the continuity and accuracy of positioning.A multi-GNSS satellite clock linear regression prediction model with short historical data is proposed.The satellite clock prediction model is used to predict the GFZ Final clock products and CNES SSR(State Space Representation)clock products and analyze the impact on real-time PPP positioning.The experimental results show that when the visible GNSS satellites whose prediction accuracy of historical data of 10 epochs is better than 0.1 ns participate in real-time kinematic PPP positioning,the 3D(three dimensions)positioning accuracy of the Final clock product can reach 0.07 m,while the 3D positioning accuracy of the SSR clock product is 0.09 m.(2)An automation offline extrinsic calibration algorithm for camera and LiDAR based on a 3D object is proposed.Aiming at the problem of the degradation of plane checkerboard pose estimation and the poor operability of camera and LiDAR calibration,an automatic extrinsic calibration method of camera and LiDAR using 3D checkerboard is proposed.A quick sorting method for the checkerboard and corner points is designed to ensure the correspondence between the checkerboard and corner points and the measurement points of the total station.Using human-computer interaction to select seed points effectively improves the problem of extracting planes from single-frame LiDAR point clouds.Experiments in the simulation environment show that that the translation error et is better than 0.9 cm and rotation error er below 0.02 rad obtained in this method.And the 3D reconstruction effect of point cloud and image fusion in the real-world environment qualitatively confirms the accuracy of this method for extrinsic calibration.(3)An offline extrinsic calibration algorithm for GNSS/IMU(Inertial Measurement Unit)and LiDAR based on a plane target is proposed.In response to the problem of GNSS/IMU positioning noise and LiDAR point cloud movement distortion and cumulative errors in LiDAR Odometer(LO)under the dynamic environment,an extrinsic calibration method for GNSS/IMU and LiDAR with the custom calibration pattern is proposed.The calibration process can be executed in a static environment and can obtain continuous and high-precision GNSS/IMU results,and no LO is required.Experiments in the simulation environment show that the translation error et is better than 1.2 cm and rotation error er below 0.0076 rad obtained in this method.The point cloud map formed by the multi-frame point cloud stitching in the real-world environment effectively eliminates the ghosting and deformation,which qualitatively confirms the accuracy of this method for extrinsic calibration.(4)A tightly coupled method for GNSS/IMU/LiDAR based on an error-state iterated kalman filter is proposed.To solve singularity of the axial angles and Eulerian angles and the truncation error of extended kalman filter for the tightly coupled LiDAR-IMU SLAM,the errorstate iterated kalman filter is used,and the combination of GNSS is the characteristics of the global positioning system and no accumulated error.A GNSS,IMU,and LiDAR tightly coupling SLAM frameworks are proposed based on the error-state iterated kalman filter.It consists of two subsystems:LIO and GNSS/IMU system.The experimental results show that this method has significantly improved the positioning accuracy compared with A-LOAM、Fast-LIO2 and the GNSS-LiDAR/IMU loosely coupled algorithm.(5)A GNSS/IMU/LiDAR/vision tightly coupled SLAM method based on graph optimization is proposed.Aiming at the complementary characteristics among vision/IMU system,LiDAR/IMU system and GNSS,a tightly coupled framework of GN S S/IMU/LiD AR/vi si on based on graph optimization is proposed,which includes GNSS/IMU system,vision/IMU and LiDAR/IMU system.To ensure the consistency of mapping,Scan Context and DBoW2 loop detection algorithms are added,and an adaptive factor real-time estimation strategy is introduced to optimize the weight of each sensor.The experimental results show that this method has significantly improved positioning accuracy compared with the GNSS-LiDAR/IMU/vision loosely coupled SLAM algorithm,LiDAR SLAM algorithm,LiDAR/IMU SLAM algorithm and Vision/IMU SLAM algorithm.After closed-loop optimization,it provides a point cloud map with global consistency while achieving global high-precision positioning.The core problems and contributions mentioned above are extensively experimentally evaluated and verified in this thesis on real-world datasets and in simulations.The experimental results show the effectiveness and superiority of the proposed theory and algorithms.The selfdeveloped handheld/backpack SLAM system integrating GNSS,IMU,LiDAR,and RGB camera and the proposed algorithm module have been successfully applied to the demonstration zone for the National Key Research and Development Program of China "Research on Key Technologies for Intelligent Land Surveying in Villages and Towns".Theory combining with practice shows significant practical values.