Abstract: To improve the accuracy of estimating sample of complex surfaces with the features such as edges and sharp corners,an estimation method for normal point clouds based on clustering classification of characteristic areas was proposed.Point clouds were hierarchically segmented into smoothness,feature boundaries,edges,sharp corners and other areas and feature types of sample points were identified by clustering analysis based on the local smoothness of surfaces and Bayesian Information Criterion.The results of normal estimation of smooth areas iteratively spread to neighbouring characteristic areas to make the normal feature points be consistent with the neighbouring smooth areas.The experiments results showed that the method could accurately estimate the normal of characteristic areas' sample points,effectively ensure the normal polysemy to the sample points of edges and sharp corners areas and suppress data noise.

Key words: normal estimation, local smoothness of surfaces, Bayesian information criterion, cluster analysis

摘要： 针对含有棱边和尖角等特征的复杂型面采样点云,为提高其法向估计结果的准确性,提出一种基于特征区域聚类分级的点云法向估计方法,根据曲面局部平坦性和贝叶斯信息准则对点云进行聚类分析,依次将点云划分为平坦、特征边缘、棱边尖角等区域,并识别样点所属特征类型,将平坦区域样点法向估计结果向其邻近特征区域依次传播,使特征样点的法向估计结果与其邻近平坦区域样点的法向保持一致。实验结果表明,该方法可以准确估计特征区域样点的法向,有效保证棱边和尖角区域样点的法向多义性,并对数据噪声具有抑制作用。

关键词: 法向估计, 曲面局部平坦度, 贝叶斯信息准则, 聚类分析