基于多尺度邻域平移法向估计算法——点云的法向量估计

A Normal Estimation Algorithm Based on Multi-Scale Neighborhood Translation—Point Cloud Normal Estimation

摘要: 本文以邻域漂移的思想为基础，提出一种简洁高效的点云法向估计算法，该算法在保证效率的同时可以有效克服边界点的法向估计效果较差的问题。首先，设计了一种多尺度候选邻域集的生成方法，该方法以当前点的所有近邻点为中心，通过多尺度的策略丰富了候选邻域集的内容，进而提高了最优邻域的质量；其次在最优邻域的评价上不仅采用协方差对候选邻域集的“平坦程度”进行刻画，而且也考虑了候选邻域与当前点的距离，从而筛选出最合理的最优邻域，提高了法向的质量。实验结果表明，该算法可以有效的克服噪声和非均匀采样等问题，很好的恢复模型的尖锐特征，更好的权衡质量和时间。

Abstract: Based on the idea of neighborhood drift, this paper proposes a simple and efficient point cloud normal estimation algorithm, which can effectively overcome the problem of poor normal estima-tion effect of boundary points while ensuring efficiency. First, a method for generating a multi-scale candidate neighborhood set is designed. This method takes all the neighbors of the current point as the center, and enriches the content of the candidate neighborhood set through multi-scale strate-gies, thereby improving the optimal neighborhood. Secondly, in the evaluation of the optimal neighborhood, not only the covariance is used to describe the “flatness” of the candidate neighbor-hood set, but also the distance between the candidate neighborhood and the current point is con-sidered, so as to screen out the most reasonable optimal neighborhood domain, which improves the quality of the normal. The experimental results show that the algorithm can effectively overcome the problems of noise and non-uniform sampling, restore the sharp features of the model well, and better trade off quality and time.