MESH MODELLING OF 3D POINT CLOUD FROM UAV IMAGES BY POINT CLASSIFICATION AND GEOMETRIC CONSTRAINTS
The point cloud generated by multiple image matching is classified as an unstructured point cloud because it is not regularly point spaced and has multiple viewpoints. The surface reconstruction technique is used to generate mesh model using unstructured point clouds. In the surface reconstruction process, it is important to calculate correct surface normals. The point cloud extracted from multi images contains position and color information of point as well as geometric information of images used in the step of point cloud generation. Thus, the surface normal estimation based on the geometric constraints is possible. However, there is a possibility that a direction of the surface normal is incorrectly estimated by noisy vertical area of the point cloud. In this paper, we propose an improved method to estimate surface normals of the vertical points within an unstructured point cloud. The proposed method detects the vertical points, adjust their normal vectors by analyzing surface normals of nearest neighbors. As a result, we have found almost all vertical points through point type classification, detected the points with wrong normal vectors and corrected the direction of the normal vectors. We compared the quality of mesh models generated with corrected surface normals and uncorrected surface normals. Result of comparison showed that our method could correct wrong surface normal successfully of vertical points and improve the quality of the mesh model.