Point Cloud Skeleton Extraction Algorithm Based on Region Segmentation

Abstract

In order to solve the problem that the L1 median skeleton extraction method involves many iterations and the skeleton easily crosses the region of the tight adjacent region,an algorithm for extracting skeleton after segmentation is proposed.According to the connectivity of the region of point cloud and the local correlation characteristics,the point cloud can be segmented into different regions using the Markov Random Field(MRF) model.Different initial contraction neighborhood scales are adaptively calculated in terms of region size and number of points in the same labeled region.The skeleton branches of each region are extracted by L1 median iteration.The skeleton connection is determined by Principal Component Analysis(PCA) and connection angle.Then the skeleton branch is connected to a complete point cloud skeleton according to the connection mode.Experimental results show that the algorithm can adaptively extract the skeleton of the points cloud and reduce the number of iterations to contract points cloud.It not only can keep the original topological structure of the model,but also has a good effect on the model with uneven region tightness.