Study On The Surface Reconstruction For Scattered Point Cloud Data

Due to an increasing need for techniques capable of processing massive,point cloud data and point cloud are receiving a growing amount of attention as arepresentation in many relative applications such as topography, computer aideddesign(CAD), medical visualization, virtual reality anddigital archive.In order to achieve better suface reconstruction effect, point cloud must bedenoise and simplification process. In the paper, we use denoising algorithmbased on k d tree, realize the denoise of unordered point cloud, the result isperfect. After that the number of the point cloud data is still very large. So itneeds further simplified to point cloud data. In the paper, we adapt retainingfeature region simplify method and study retaining sharp edges of the data whichto simplified to data in smooth region, this method complete retains the data ofshap edges and the shape of the smooth region.In the above basis, this thesis researchs a novel unorganized point cloudreconstruction scheme based on the mean shift clustering approach. This schemedoes not request the users to input or tune any parameters (they are determinedautomatically). This scheme only takes the point cloud data as inputs.There are three main steps in the reconstruction process. First, the pointcloud is decomposed into several near flat and disk like patches by utilizing themean shift clustering approach. Second, each separated patches areparameterized on2D disk using local parameterization approach. Third, adoptDelaunay triangulation to construct the connectivity of surface patch.Eventually,the triangle mesh of point cloud is obtained by merging all separated patches.Therefore, the surface reconstruction can convert to simple two dimensionalerea of triangulation. the method has been verified and it easily solute oversmooth of feature regioonal and achieve ideal reconstruction.