Research On Multi-view Automatic Registration Algorithm For Terrestrial Laser Scanning Dat

Terrestrial laser scanning(TLS)is a non-contact active and fast acquisition of three-dimensional dense point clouds on the surface of objects.Its technical principle is to use light in the form of pulsed lasers to measure the distance from the sensor to the object.Ground 3D laser scanning technology is an important means of 3D digitization and reconstruction of ground objects.It has the characteristics of convenient portability,low cost,direct acquisition of 3D spatial information of ground objects,high precision,and high density.It has been used in building digitization,heritage protection,terrain monitoring,Indoor modeling and other aspects have been widely used.Some commonly used dual-view and multi-view registration algorithms currently have problems such as easy to fall into local minimum,poor registration effect when the initial pose is not good,slow registration algorithm speed and low accuracy in large-scale measured point clouds.In response to these problems,this paper conducts research based on the measured multi-view ground laser scanning data.The innovations and main work are as follows:(1)In view of the uneven density of ground laser scanning data points,this paper improves the radius filtering algorithm,and adds distance weights on the basis of the radius filtering algorithm.This method first uses the kd tree to search for the number of neighbors in the neighborhood of the target point;then sets different filtering conditions and thresholds according to the range of the target scene and the distance between the sampling point and the sensor;finally,the number of neighbors and the threshold In comparison,if it is less than the threshold,the sampling point is considered to be an outlier and should be deleted.Experiments show that compared with traditional statistical filtering algorithms,this algorithm can better remove outliers and retain point clouds with high local geometric features.(2)This paper uses an improved two-view registration method based on statistical features of rotation projection.This algorithm takes advantage of the high descriptiveness and high robustness of the Ro PS descriptor,and uses the threedimensional local reference coordinate system constructed by the weighted covariance matrix to replace the triangulated point cloud grid required in the original Ro PS descriptor,which solves the problem.Descriptor construction is difficult and inefficient.The comparison experiments with commonly used registration algorithms such as NDT,SAC＿IA,4PCS,etc.prove that the algorithm has high noise resistance and accuracy.(3)In view of the characteristics of highly repetitive ground laser scanning data,this paper uses a dual-view registration algorithm that uses the automatic extraction of target spheres.This algorithm is suitable for point clouds that lack geometric spatial feature differences such as corridors and tunnels.The algorithm first uses the octree to segment the point cloud to improve the calculation efficiency;then uses the RANSAC algorithm to eliminate the plane points and low curvature surface points in the point cloud,and completes the preliminary extraction after clustering;then uses the method based on probability statistics to extract the spherical surface and Fit the center point;finally use the center point to construct a similar triangle feature to find the matching point with the same name,and complete the registration based on the matching point with the same name.Experimental results show that the algorithm can accurately extract the position of the target ball and successfully complete the registration task.(4)The current traditional multi-view point cloud registration algorithm only uses the exhaustive method for pairwise sequential registration of point clouds that contain overlapping parts.This method is quite time-consuming and has high computational complexity,and cannot be used for large-scale point clouds.registration.Aiming at this shortcoming,this paper optimizes the hierarchical multiview registration method based on the maximum spanning tree.This method first uses a compound distance to calculate the overlap between point clouds;then according to the obtained overlap,constructs the maximum The spanning tree restores the best registration order between point clouds,and finally the hierarchical registration is merged continuously,and finally unified into the same coordinate system.Based on the ground 3D laser scanning data,this paper improves the traditional algorithm in point cloud preprocessing,dual-view point cloud registration and multiview point cloud registration,which is conducive to improving the performance and practicality of automatic registration of 3D ground laser scanning data.sex.