Research On Multi-source Data Fusion Algorithm For 3D Modeling

Three-dimensional modeling is widely used in the fields of smart city management,historical and cultural heritage protection and geological disaster emergency rescue.With the development of surveying and mapping science and technology and equipment manufacturing,the three-dimensional data fusion modeling industry based on multi-source data is showing a prosperous situation.Tilt photogrammetry and three-dimensional laser scanning are common spatial data acquisition methods,which are increasingly used in production.In the study of multi-source data fusion algorithms for 3D modeling,due to the limitations of multi-source data such as heterogeneous heterogeneity and large data volume,the multi-source data fusion algorithm is inefficient and has no effect in the 3D modeling process good.Similar problems exist in data fusion based on tilt photogrammetry and3 D laser scanning acquisition.In view of the above problems,this paper conducts in-depth research on key algorithms such as image fusion processing,point cloud fusion processing and multi-source data fusion.The main research content and research results of the paper are as follows:(1)In order to solve the problem of insensitivity of conventional image edge detection algorithms in image fusion research,this paper builds an improved Laplacian edge detection algorithm.This algorithm first enhances the image and smoothes the image with Gaussian filter to suppress noise;Then the three phase components of RGB are respectively sharpened and stretched by the Laplace template;finally,the three components are phase-recombined to realize the edge detection of the improved algorithm.The thesis conducted a comparative experimental study on the improved algorithm constructed,and the research results show that the algorithm constructed in this paper has better edge detection effect than other algorithms,and the obtained peak signal-to-noise ratio(PSNR)is higher than the conventional algorithm,and the mean square The root error(MSE)is smaller,and at the same time,the algorithm has better robustness.(2)Aiming at the problems of low efficiency of classic SAC-IA algorithm in point cloud data fusion processing,slow iterative convergence speed of classic ICP algorithm,and iterative solutions generally falling into local optimal solution,this paper builds a combination of SAC based on SAC-IA and ICP algorithm-Improved algorithm of point cloud fusion processing of IA and ICP algorithms,providing algorithm support for subsequent multi-source point cloud data fusion.The processing procedure of the improved algorithm is as follows: At the initial registration,the point cloud data is simplified by using the SAC-IA algorithm,and the point cloud that does not have to participate in subsequent registration operations is deleted to improve the efficiency of the point cloud coarse registration processing;When using ICP algorithm for accurate registration,R-tree index is used to accelerate the corresponding point pair query,and then use the angle threshold of the direction vector to eliminate the wrong matching point pair,thereby improving the precision registration efficiency.The paper conducts an experimental comparative analysis of the improved algorithm constructed.The experimental results show that the improved algorithm is superior to the classic algorithm in terms of time-consuming,mean square error,and registration error.The efficiency has improved significantly.(3)Aiming at the problem of incompatibility of fusion data in different data sources,especially the oblique image point cloud and lidar point cloud in multi-source data fusion,this paper constructed an R-tree indexed point cloud fusion processing algorithm to restore motion The dense point cloud generated by the structure(SFM)performs the consistent conversion of the data structure and the model,and provides technical support for the subsequent two data fusion modeling.The algorithm first creates a three-dimensional R-tree index on the point cloud and loads and saves it to the point cloud data;then it combines the multi-view point cloud,searches the point cloud sphere neighborhood,maps the point cloud to a higher dimensional space,and establishes the largest spaced hyperplane in the space On this basis,multiple parallel hyperplanes are used to separate point clouds,and triangles are generated using the Delaunay growth method,and then point cloud data structure matching identification is performed;finally,linear conversion is performed.(4)The thesis is based on oblique photography point cloud data and lidar pointcloud data,fused two kinds of data and carried out 3D modeling experiments,and compared with the results of 3D modeling from a single data source.Experimental results show that the modeling effect based on multi-source data fusion is better than that of a single data source.