Research On Truck Point Cloud Registration Method Based On Lidar

With the rapid development of the logistics industry,the traditional manual loading method has limited the improvement of transportation efficiency.The truck is scanned by LIDAR to build a 3D model to achieve truck attitude positioning and body segmentation,providing digital support for automatic loading.LIDAR has a limited viewing angle and the laser is easily blocked,requiring multiple scans and alignments from multiple angles to obtain a complete point cloud of the truck.Therefore,point cloud alignment is a key part of building 3D models of trucks and automating loading.In this thesis,a dual LIDAR scanning system suspended on both sides of the loading bay is designed to achieve full truck coverage scanning.For the noise at different scales present in the original point cloud data,a point cloud denoising method based on Laplace kernel is proposed to fuse the Laplace kernel density estimation function to evaluate the influence of points in their neighborhoods,so as to reject the noise;To address the point cloud alignment problem of unilateral scanning system,this thesis proposes a homologous point cloud alignment method based on the combination of IMU and improved ICP,which introduces the inertial measurement unit into the point cloud alignment,generates a global transformation matrix for each input point cloud to complete the point cloud position correction without relying on the point cloud data,and then improve the ICP algorithm and incorporate the random sampling algorithm to fit the local plane to search for the nearest point to complete the accurate alignment of the point cloud,which improves the alignment efficiency and accuracy;The point cloud on the side of the truck near the scanning system is sparse,and the point cloud model of the truck on both sides needs to be aligned again.This thesis propose a KD-Tree based density clustering(DBSCAN)heterogenous point cloud alignment method.Designing alignment targets,reorganizing the point cloud space with the KD-Tree method,optimizing the neighborhood construction accuracy,improving efficiency,and incorporating the DBSCAN algorithm to improve the target fitting accuracy;The calculated conversion matrix is used as the initial value of the improved ICP method,which avoids the ICP method from falling into local optimum and improves the alignment accuracy.The method of this paper was evaluated by collecting truck point cloud data with a dual Li DAR scanning system.The single-side homologous point cloud alignment can be completed within 4s,with the maximum translation error not exceeding 0.01 m and the rotation error controlled within 0.1°;for the two-side heterogenous point cloud alignment,the average rotation error is controlled within 0.1° and the average translation error is controlled within 0.02 m,showing good performance.