摘要
点云法矢计算对点云分布密度较敏感,而且在尖锐边界处计算误差较大。为此,提出一种基于自组织神经网络(SOM)的散乱点云法矢计算方法。为利用散乱点云拓扑和几何信息计算法矢,以球面SOM学习点云拓扑结构,得到被测曲面的三角网格近似图,使用三角网格构成的连通图组织点云数据结构,通过k-近邻点拟合微切平面,从而计算点云法矢,并调整点云法矢指向。实验结果表明,该方法具有较高的计算精度,法矢误差在0.08以内,标准差为0.009。
The point cloud normal vector calculation is sensitive to distribution density, and the calculation error is big in sharp border presently. In order to solve this problem a method of normal vector calculation based on Self Organization Map（SOM） is presented. The geometrical and topological information on scattered point cloud are employed to estimate the normal vector. A sphere SOM is trained to approximate the sampled surface with triangular meshes. Point cloud is clustered on the nodes of SOM, after that plane fitted by the k-neighbor points gives an estimation of the point normal. And the estimated point cloud normal vectors are aligned by adjusting patch normal. Experimental results show that the relative error is less than 0.08 and the standard deviation is 0.009. The method has high calculation precision.
出处
《计算机工程》
CAS
CSCD
2012年第8期287-290,共4页
Computer Engineering
基金
广东省高校优秀青年创新人才培养计划基金资助项目(LYM10121)
关键词
散乱点
拓扑信息
自组织神经网络
法矢计算
逆向工程
scattered point
topological information
Self Organization Map（SOM）
normal vector calculation
reverse engineering
