Research On 2-manifold Surface Construction Based On Graph Cut

In recent years,with the vigorous support of the state and the construction of "smart city" in big cities,the technology of image data acquisition by remote sensing has also developed rapidly,especially for the UAV platform for large-scale image data acquisition technology in low-altitude cities.Because of the rapid development of image data acquisition technology,image data has the characteristics of wide range,many objects and miscellaneous data,which brings great challenges to the traditional three-dimensional networking technology.The traditional projection method can not express the three-dimensional surface of buildings,and the mainstream three-dimensional meshing algorithm can not adapt to the actual large-scale scene image data in terms of both model quality and processing efficiency.Therefore,it is of great significance to find a three-dimensional network construction method suitable for real large scene images.In view of the above problems,this paper analyses the three-dimensional network construction algorithm on the basis of both efficiency and quality,and chooses the graph cut method with better overall performance to carry out the three-dimensional network construction experiment.At the same time,aiming at the problem that the surface constructed by graph cut method is non-manifold,an improved strategy of constructing manifold surface based on S-T icon label redistribution is proposed,which brings convenience to the following triangular mesh simplification work.The main research results are as follows:(1)For the dense point cloud data needed by graph cut method,this paper analyzed the sparse point cloud reconstruction algorithm based on motion recovery structure(SFM),studied the dense point cloud reconstruction algorithm based on matter-aspect element(PMVS)multi-view,and used the Bundle system based on SFM and PMVS combined reconstruction strategy to reconstruct dense point cloud image data,and obtained dense point cloud.The dense point cloud information provides a good data support for the experiment of three-dimensional network construction by graph cut method.(2)The Graph Segmentation Algorithm Based on ray consistency segmentation is studied.Firstly,three-dimensional Delaunay tetrahedron is constructed from dense point cloud data.On the basis of tracing the visible light passing through Delaunay triangle,a weight value of these triangles is given.According to the relationship between Delaunay tetrahedron and Voronoi polygon,the S-T graph is established.The Delaunay triangle is segmented by the principle of maximum flow and minimum cut,and the reconstructed surface is obtained.Thirdly,aiming at the problem that the surface generated by traditional graph cut method is non-manifold,the causes of non-manifold and non-manifold phenomena are summarized.An improved strategy of S-T icon label redistribution is proposed,and the feasibility of constructing manifold surface by graph cut method is verified by networking experiments.(3)For triangular mesh simplification technology,the edge collapse simplification algorithm suitable for manifold surface is studied.The improved graph cut method is simplified to reconstruct the surface,and a better simplified structure is obtained.At the same time,different end-collapse thresholds are set.The feasibility of constructing manifold surface by graph cut method is further verified by comparing and analyzing the results of simplified mesh.In this paper,the feasibility of constructing manifold surface based on graph cut method is verified by experiments.At the same time,the experiment uses Bundle+PMVS to reconstruct dense point clouds,uses improved graph cut method to simplify the edge collapse of reconstructed triangular grid,and finally gets a set of processing strategies of three-dimensional model through texture mapping,which provides a set of feasible processing ideas for three-dimensional network construction technology.