Research On Feature Skeleton Extraction Based On Point Cloud

The 1D curve skeleton can intuitively represent the geometric structure and topology connectivity of the 3D point cloud.The curve skeleton information can be used to analyze and manipulate various graphic data.It is widely used in skeleton matching and analysis,skeleton modeling,skeleton animation and point cloud restoration and other fields.However,there are still many problems in the extraction of skeleton features for 3D point cloud data:The initial point cloud model obtained by scanning often suffers from a lot of noise and large area of cloud points because of the technical and environmental constraints.Because the point cloud model does not contain explicit topological connection information,it is very difficult to extract the skeleton of the point cloud model.At present,the mainstream point cloud skeleton extraction algorithm needs to adjust a large number of parameters to obtain an accurate skeleton,resulting in a decrease in the practicality of the algorithm.In order to meet the actual needs,a practical and effective 3D skeleton model is proposed.In this paper,the related technology of the curve skeleton extraction of the scanning point cloud model is studied,and a method of extracting the distance field guided ₁L-median skeleton is proposed.The method first voxelizes the 3D point cloud model uniformly by using a stepwise subdivision method;Second,using the fast matching method to calculate the distance field of the interior voxel points of the model;Then use the distance field to extract the initial skeleton of the model using the multi-scale parameter control refinement method;Finally,we incorporate the initial skeleton into the ₁L-median skeleton extraction algorithm and generate the distance field basedL₁-median skeleton extraction to extract a complete skeleton from the point cloud model quickly and accurately.In order to verify the validity of this research method,we use the method of this paper to extract the curve skeletons for different complicated shape scanning point cloud models.The experimental results show that this method can not only deal with the point cloud model with a large number of noise points,outliers and point clouds,but also can calculate a more accurate 1D curve skeleton within two minutes for the point cloud model with less than forty thousand points.At the same time,the method does not produce wrong skeleton connection,nor does it require too many parameters.