A Thinning Algorithm for Topologically Correct 3d Surface Reconstruction

The existing algorithms for 3D surface reconstruction from point clouds require objects with smooth surfaces and extremely dense samplings with no or only very few noise in order to guarantee a correct result. Moreover they can only be applied to single object reconstruction and not to segmentation of three or more regions. We have developed an alternative surface reconstruction method wich preserves the topological structure of multi-region objects under much weaker constraints. It is based on the Delaunay complex and α-shapes and uses a local thinning algorithm for the reconstruction of region boundaries. In this work we give a detailed analysis of its behaviour and we compare it with other approaches.