Heterogeneous Point Cloud Registration Method Based On Common Line And Surface Features

Point cloud,as a crucial data representation in three-dimensional space,exhibits characteristics of "large volume" and "multisource" amidst the rapid development of sensor technology.Point cloud data acquired from different collection devices and methods possess their respective advantages and limitations.For instance,laser scannergenerated point clouds offer accurate three-dimensional coordinates,realistic scale features,and intensity information,yet they often suffer from sparsity and color information loss.On the other hand,image-based matching point clouds exhibit high density,authentic color information,and edge feature details,but they are prone to the occurrence of "point cloud holes." By fusing point clouds from multiple sources,the deficiencies of individual point cloud data sources can be overcome,leading to complementary advantages and a more comprehensive and holistic depiction of geographical entities.Point cloud registration is one of the key approaches to achieving the fusion of multisource point clouds.Based on this,this thesis studies the registration method of heterogeneous point clouds based on common line and surface characteristics,and realizes the registration fusion of Lidar point cloud and image matching point cloud.The registration accuracy can reach decimeter level,which has certain reference significance for the research of multi-source point cloud data fusion.The main research contents are as follows:(1)This thesis analyzes the differences between Lidar point cloud and image matching point cloud in terms of acquisition methods,point cloud density,data accuracy,and other factors.Based on their respective advantages and disadvantages,the fusion demand between point cloud data is analyzed,and a feasible registration scheme for different source point clouds is proposed based on the fusion demand.(2)Based on improved RANSAC plane fitting method.In order to solve the threshold dependence problem of RANSAC algorithm in plane fitting,an improved strategy of "dynamic threshold" is proposed,and the distance threshold is automatically set and updated under the constraint of the expected proportion evaluation factor.Comparative experiments show that this method can effectively overcome the influence of point cloud noise and outliers and better fit the plane features.At the same time,this thesis uses the method of intersection line of the extracted common planar features instead of the manual selection of linear features.(3)Point cloud registration method based on common line and surface features.The mathematical model of coordinate transformation based on common line and common surface features is studied,and the transformation model is applied to point cloud registration.Constraints are established through the extracted common line and surface features,and the transformation matrix is solved.The transformation matrix is applied to the source point cloud to complete the registration of different source point clouds,and the accuracy of the registration results is evaluated.Through different source point cloud registration experiments in different scenarios,the feasibility and practicability of the registration method based on common line and surface feature point cloud are verified.