High Precision 3D Environment Reconstruction Based On LiDAR Point Cloud Plane Detection

Currently,virtual simulation-based "meta-verse" and "digital twin" technologies have become hot topics in the field of 3D reconstruction.The reconstructed 3D scenes can significantly increase the fidelity of the environment and provide users with a more immersive and realistic experience.Based on the principle of laser ranging,Li DAR sensors have the advantages of high speed,high precision,and intuitive information of 3D environment scanning.The Li DAR has widely used in the fields of 3D reconstruction and autonomous positioning.However,due to the limitation of vertical resolution,it is necessary for Li DAR to measure from multiple directions to obtain 3D point cloud data of the environment in all directions.To unify the point cloud coordinates from different perspectives into a global coordinate system and achieve high-precision scene reconstruction,this study completed the following works:(1)A monochrome triangular pyramid is designed as a scene calibration object to assist in obtaining 3D point cloud data in all directions,thus compensating for the scanning defects of the vertical field of view.At the same time,the calibration object is used to associate multiple frames of point cloud scenes to improve the algorithm’s scene adaptability and reconstruction accuracy.Additionally,the planes of calibration object,formed by fitting a large number of point clouds,is used for feature extraction,which makes the detected spatial point cloud features more stable.(2)To improve plane detection accuracy,a combination of the RANSAC(Random Sample Consensus)algorithm and the Least Square Method is used to fit the point cloud plane.Furthermore,the Three-axis Least Square Method(3A-LSM)algorithm is proposed for theoretical innovation to further reduce the fitting error of3 D planes in the point cloud space,thus ensuring the accuracy of subsequent point cloud scene feature extraction.(3)The virtual corner points generated by the intersection of multiple planes in space are proposed as scene feature descriptors,improving the stability and robustness of point cloud scene feature extraction.Additionally,a small number of virtual corner points are used as input for the Iterative Closest Point(ICP)algorithm,which performs scan matching based on local features,effectively improving the real-time performance of system and reducing the time and memory consumption of point cloud registration.Furthermore,adjacent frame registration is used to solve the problem of ICP requiring a good initial value,avoiding the model converging to a local minimum.(4)A loop closure optimization module is added to the 3D reconstruction system,which associates single-frame scenes with the global scene,constraining the pose of each frame scene based on the feature points in the global scene,and reducing the cumulative error generated during scene fusion.After loop closure optimization,the average distance between points and the calibration plane reaches2.75 mm in the global reconstructed scene,and the reconstruction accuracy is improved by 12% compared to before loop closure optimization,meeting the high-precision 3D reconstruction point cloud model requirements of current enterprise engineering.