The Research On Skeleton Extraction Of Three-dimensional Point Cloud

With the improvement of computer performance and three-dimensional scanning technology matures.As a representation method of object shape information Three-dimensional point cloud is more and more used in engineering design,games,animation and other research areas.The 3D point cloud skeleton can express the shape topology information of the model and more convenient to interact.It can also improve the storage and computational efficiency of the model.Therefore,it is one of the important research questions to extract high quality skeleton for 3D point cloud data.The skeleton is a simple structure representation of the three-dimensional model.It can reflect the basic topology of the three-dimensional model and shape,while ignoring the redundant information contained in the original model.There are two types of skeletons:One is the curve model,called a curve skeleton.The other is the central axis model.According to different practical needs,can be simplified the skeleton.The most commonly used is the simplified skeleton.This model can reflect the model of the topology,but also have a more refined form of expression.So the curve skeleton is more commonly used to express the topology of model.A robust method for three-dimensional point cloud skeleton extraction is proposed in this paper.Firstly,Laplace operator is used for contracting three-dimensional point cloud locally,and principal component analysis(PCA)is performed on the contracted point cloud to extract the skeleton branches.Then,the local point that has been extracted skeleton will be fixed and the rest will have further contraction.The point cloud contraction and local skeleton extraction are repeated until obtain a complete skeleton curve which satisfies the termination conditions.Finally,by processing the cross points of skeleton and cubic B-spline curves fitting,we can get the final skeleton curves.In this paper,we have improved the method of skeleton extraction from the following three aspects:(1)Based on the Laplace operator,we propose a local point cloud contraction method.Our method can effective to prevent local point cloud data contraction,and the rest of the point cloud data contraction is not enough,while can't extract the skeleton.(2)In view of the local point cloud contracted method,a skeleton extraction method based on dynamic PC A computation is proposed.When extracting the local point cloud skeleton,it can dynamically adjust the calculated radius of the PCA,automatically calculate the point cloud skeleton branch,and finally merge the skeleton branches.(3)The extracted skeleton was classified by the topology and was optimized by B-spline curve.Experimental results show that compared with other existing methods,point cloud skeleton extracted by the proposed method is more robust and resistant to noise.