Research And Implementation Of Point Cloud Model Of Skeleton Extraction Algorithm

Extraction of curve-skeletons is a fundamental problem with many applications in computer graphics and visualization. Curve-skeletons are1D representation of3D objeets, which are useful commonly for many visualization tasks including educational virtual realization, educational model searching, educational game, etc. Most existing curve-skeleton extraction methods make use of a volumetric diserete represeniation or mesh surface representation. But methods for extraction curve-skeletons from point clouds are seldom relatively.Due to recent advances in high-speed3D laser-scanning technologies, a typical point cloud contains millions of coordinate data points which lead to significant challenges for the following curve-skeleton extraction. In this paper we propose a simplification method based on Quadric Error Metric (QEM) for point clouds. It firstly smoothes the input point cloud to remove the outliers and noise based on Quadric Error Metric. After smoothing, feature points of the smoothed point cloud are extracted depending on local curvatures. And some measures have been taken to protect the feature points from being removed. Then the radiuses and the optimal points of covering balls which are centered at the non-feature points are computed through the method of Quadric Error Metric (QEM). At last, the optimal points are set to replace the non-feature points inside the covering balls. In this method, not only the non-feature points which contribute less are simplified, but also the geometric features of a original point cloud are reserved.Redundancies have been removed after above simplification. At the same time the data missing and noise may exist in the point clouds due to the complexities of the original models. And the multi-branches of a model may interfere with each other in skeleton extraction. In view of the above problems, this paper proposes a robust approach to extract curve skeletons from incomplete point cloud even with noise and outliers based on point cloud segmentation. Our method firstly segments the input point cloud models and then extracts skeleton points based on the notion of generalized rotational symmetry axis (ROSA). After that the skeleton points are thinned through filtering. At last, one dimension curve skeleton extracted by fitting the ordered skeleton points obtained by down-sampling. Our approach could avoid the interferences between multi-branches and be robust to deal with challenging point clouds with holes and noise which are meaningful for following geometric processing.