Research On Surface Reconstruction Based On Optimization Of Surface Local Sample

Rencently, the accurate reconstruction of sharp feature has been the focus of domestic and international research in the field of surface reconstruction. The surface local sample used in current surface reconstruction methods has bias, which could influence the accuracy of the reconstruction results. The unreasonable reconstruction methods also cause defects such as holes, non manifolds and hollows in reconstruction results. In this paper,the surface local sample are optimized with gaining more valid points, and the optimized surface local samples are used in the classical reconstruction methods, α-shape and Cocone,to create new algorithms for improving the accuracy of the sharp feature reconstruction results.The headlines and results are list below:(1) The k-neighbours of a sample point is obtained with the KD tree structure,and used as initial surface local sample of the corresponding region at the physical surface.The probability density of the points in the surface local sample is estimated by kernel density estimation method,then the K-means algorithm is used to cluster the points in local sample to find the data search direction, and then some points in the sparse region are added into the local sample.The above iterative procedure could be seen as gain optimization of the surface local sample,and it efficiently makes the local sample better adapted to the point distribution at local region of the physical surface.(2) To optimize the α-shape algorithm, our algorithm uses approximation of the topological neighbors of a sample point as the surface local sample, which makes α-shape scale thresholds reflect density of the points better, then the validity of the surface reconstruction is improved. Gaining the approximation of the topological neighbors of a sample point is essentially to achieve gain optimization for the Euclidean neighbors of the point, which extends the latter toward the sparse region of the sampled data so that it decreeses dropping of the topological neighbors caused by non-uniform points. Based on the approximation of the topological neighbors of points and prior knowledge of surface reconstruction, an α-shape scale threshold corresponding to an triangular face could be calculated, so that the scale thresholds used in surface reconstruction could be adjusted by itself adaptively. The tests show that this algorithm can reconstruct non-uniform point set with few holes and edge hollows, better maintain the accuracy of form and position, and reduce non-manifold facets, meanwhile, its efficiency is comparable with mainstream algorithms.(3) A sharp feature adapted Cocone algorithm is proposed,which could improve the reconstruction accuracy of sharp feature by revising the normal vector estimated at points and making the Cocone angle thresholds adapted the points ant sharp feature region.The local sampe used by PCA normal estimation is optimized by gain optimization method,then the normal estimation result could be revised and become more adapted to sharp feature region.Boundary points are identified by comparing the deviation extent between the objective point and its mode points,then a self adapted Discrete Gauss clustering is used to identify the sharp feature points from the boudarty points.A Cocone angle threshold adjustment criterion is created to make the Cocone angle thresholds corresponding to sharp feature points more adapted.