保留几何特征的散乱点云简化算法

摘要

In this paper, we proposed a robust point cloud simplification method which can hold the geometric features and the shape of the potential surface. First, by taking the normal change within unit distance as the local feature operator, Poisson region growing method was employed to detect feature points and calculate the average curving degree of the potential surface. Second, by setting the clustering thresholds adaptively, the shared nearest neighbor (SNN) clustering method was utilized to analyze the neighborhoods of non-feature points, measure the local curving degree of the potential surface, and detect the outliers. Finally, all of the outliers were removed and the feature points were kept. Furthermore, in order to retain the curving degree of the potential surface, avoid holes, and decrease the simplification error, different simplification strategies are adopted for different non-feature points. The experimental results presented in this paper demonstrate that, the proposed simple and ef-fective method can not only retain both the features and the curving degree of the potential surface, but decrease the simplification error as well.%针对散乱点云简化中易丢失几何特征及潜在曲面形状信息的问题，提出一种保留几何特征的散乱点云简化算法。首先以单位距离上的法向变化作为局部特征检测算子，采用基于泊松分布的区域生长法自适应地检测特征点，并计算潜在曲面的平均弯曲度；然后通过设定不同的聚类阈值，并利用共享近邻聚类算法对非特征点的邻域进行聚类分析，从而判定该点处潜在曲面的弯曲程度，同时检测噪声点；最后，删除噪声点，根据潜在曲面弯曲程度，采用不同的简化策略删除冗余点。该算法不但避免了在大量精简时造成孔洞，而且使得简化后模型尽可能保持原始潜在曲面的形状信息，降低简化误差。实验结果表明，文中算法简单、有效，能够同时保留原始点云的几何特征及潜在曲面的形状信息，具有较低的简化误差和良好的鲁棒性。