| Most cultural relics protection units use 3D laser scanning technology,photogrammetry and other non-contact measurement methods to digitally record the status quo of cultural relics,which has become an important means of high-precision 3D reconstruction of precious cultural relics.The 3D models obtained by these two technologies have unavoidable defects.The texture data of the laser scanning model is distorted,and the accuracy of each point of the camera measurement model is inconsistent.In order to build a high-precision geometric and fine-textured model,this paper fuses and reconstructs the data of these two non-homologous models according to their respective advantages,and combines the theory and foundation of Web GL to build a visualization platform for digital 3D models based on the open source engine Cesium.Management and visualization,the main research contents are as follows:(1)Aiming at the problem of fusing the first two model data in different coordinate systems,this paper proposes an iterative improvement of the nearest neighbor algorithm based on the octree grid to improve the laser scanning point cloud data and image point cloud data from coarse to fine and efficient.High-precision registration method,through the improved SAC-IA coarse registration algorithm and the improved ICP algorithm,the high-precision registration of non-homogeneous data of scanning point cloud and image point cloud is realized.The experimental results show that: on the premise of not losing the overall contour features,the grid-based down-sampling of two sets of point cloud data reduces the complexity of feature description calculations.Compared with the 4PCS algorithm and RANSAC algorithm,this paper proposes an improved algorithm The registration time was reduced by at least 13% and 28%,respectively,and the registration accuracy was increased by at least 23%and 16%.In addition,the Euclidean distance threshold and the grid were used to select the corresponding point pairs to maximize the accuracy of the two data.Registration accuracy,compared with the classic ICP algorithm,the registration accuracy of the improved ICP algorithm proposed in this paper is at least five times higher,and the registration time is only one-third of the classic ICP.(2)Aiming at the problem that the scanning model and the image model belong to nonhomogeneous and heterogeneous data,and the data of these two models cannot be directly fused,a solution is proposed in this paper: firstly,the point cloud of the image model is upsampled,and the The density of the point cloud,and after interpolating the image model point cloud,it is also necessary to perform the same interpolation processing on the corresponding texture patch,calculate the interpolation coordinates,re-divide the triangle patch,and then upsample The point cloud is used for normal estimation,and the intersection point with the scanning model is obtained,and finally texture mapping is performed to generate a new fusion model.The experimental results show that the fusion model has inherited the advantages of the scanning model and the image model,and has both good geometric precision and good texture precision.(3)In view of the problem that the existing model data cannot be directly rendered and visualized on the Web,this paper proposes a method for constructing 3D Tiles data,and optimizes the scheduling and rendering of the Web-side visualization platform.Through the detailed analysis of the Cesium architecture and the data organization of 3D Tiles,the grid is used to sample the point cloud data and the edge folding algorithm is used to simplify the triangulation model,and the 3D Tiles data with LOD structure is constructed.To meet the needs of data visualization,the overall architecture of the visualization platform is designed,and the functions of user interaction such as map display,model data loading and rendering,and data analysis are realized on the visualization platform.The experimental results show that the visualization platform has a good effect on the loading and rendering of point cloud data,color triangulation data,and tilted model data.The various data analysis functions provided by the platform are easy to operate and strong in practicability,which can meet the visualization needs of users. |
