Manifold mesh reconstruction from dense matching points of oblique images

Abstract

In the process of 3D reconstruction, constructing mesh models from matching points is a complex problem because of noise and numerous holes. For Multi-view reconstruction, it is difficult to construct triangular meshes due to the complex geometric shapes of urban scenes. Many traditional mesh reconstruction methods cannot effectively handle noise and generate surfaces with holes and the generated mesh cannot meet the characteristics of 2-manifold. In this paper, a method of constructing 2-manifold meshes based on the graph cuts is proposed. Our algorithm consists of four main steps. Firstly, constructs a directed graph by using tetrahedra constructed from point clouds. Then, weights are added to the edges and nodes based on their surface-visibility and smoothness. Thirdly, the tetrahedra are labelled as inside or outside of the surface using the graph cuts algorithm, and the labelled tetrahedra are adjusted by clustering and relabelling in the s-t graph. Finally, extract the surface from the labelled tetrahedra and represent it as a triangular mesh. The experiments show that the constructed mesh satisfies the 2-manifold characteristics and the half-edge data structure, which is convenient for the subsequent steps. Compared with APSS, IMLS and Poisson reconstruction algorithm, our method has better robustness and effectiveness.

Keywords

• UAV photogrammetry
• oblique images
• mesh reconstruction
• 2-manifold mesh
• 3D models