Research On 3D Reconstruction Method Of Mobile Robot Based On LiDAR-Visual-Inertial Fusion

There are increasing application demands for 3D reconstruction technology in architectural surveying,smart city,virtual reality,and other fields.How to quickly and effectively carry out 3D modeling has become a research hotspot.Aiming at the limitation of traditional 3D scanning technology and oblique photography technology,this paper studies a 3D reconstruction method for mobile robots based on SLAM(Simultaneous Localization and Mapping),and proposes a SLAM algorithm fusing LiDAR(Light Detection and Ranging),stereo camera and IMU(Inertial Measurement Unit),which improves the localizing accuracy and mapping effect of mobile robots in complex architectural environments,and is applied to 3D reconstruction tasks in the field of architectural surveying and mapping.The main research work is summarized as follows:(1)This paper builds the multi-sensor fusion system platform of the mobile robot,and analyzes the observation model of the sensor.The parameters are further obtained according to the principle of internal and external parameter calibration of the sensor,and the influence of the noise of the original data on the multi-sensor information fusion localizing is reduced.(2)Aiming at the problem of information redundancy in the original point cloud of LiDAR,this paper proposes a feature extraction method based on the distribution of geometric intensity information of point cloud,and uses the Scan-to-Map method to improve the matching speed between frames,and estimated the LiDAR frame pose by optimizing the geometric information residual and intensity information residual of the feature point clouds.For the stereo image information,this paper uses the optical flow method to extract and track feature points,and correlates it with the LiDAR point cloud depth to improve the accuracy of map points,and then uses sliding window optimization of visual feature reprojection residual and IMU pre-integration residual to solve image frame pose.(3)Aiming at the problem of accumulated drift when the odometer runs for a long time,this paper proposes a closed-loop detection method of point cloud ISC(Intensity Scan Context)descriptor based on geometric information,which improves the efficiency of closed-loop retrieval.The back-end builds a global factor graph consisting of laser odometry,visual odometry,closed-loop constraints and IMU preintegration errors.The key frame pose is optimized by incremental smoothing and map construction,reducing the cumulative error of the odometer and outputting a highprecision point cloud map.(4)This paper uses public datasets and measured datasets to evaluate the performance of the algorithm in this paper,and compares it with LeGO-LOAM,FastLIO and VINS-Fusion.The results show that the algorithm in this paper has higher localizing accuracy and better mapping effect,which proves that multi-sensor fusion improves the robustness of the algorithm in complex environments.Finally,the threedimensional modeling of the point cloud map output by the algorithm in this paper in the measured data is carried out,and the accuracy of the 3D model is evaluated,which shows the practical application value of this system in the field of architectural surveying and mapping.