Abstract: Point clouds are becoming more and more common for the representation of 3D geometry models because of its advantages over mesh models，such as easy acquisition，straightforward representation and flexibility.Since normal is one of the essential properties of point clouds，the estimation of normal plays an important role in point clouds processing.However，point clouds are prone to contain noise，outliers and holes because of the unavoidable noise，physical errors and occlusions during acquisition.Moreover，some point clouds，such as data from CAD models，also contain sharp features.These factors pose different challenges to normal estimation.A comprehensive survey of the recent work in normal estimation is presented，and the principles as well as key techniques of them are discussed with emphases on the ability of dealing with noise，outliers，sharp features，etc.In the end，the conclusions and the future research trends of this topic are given.

Key words: 3D point clouds, normal estimation, noise, outlier, sharp features

摘要： 由于获取方便、表示简单、灵活等优势，点云逐渐成为常用的三维模型表示方法之一。法向量作为点云必不可少的属性之一，其估计方法在点云处理中具有重要的位置。另一方面，由于点云获取过程中不可避免的噪声、误差和遮挡，点云中通常含有噪声、外点和空洞，并且部分采样模型如CAD模型，也会存在尖锐特征，这些都给法向量估计提出了挑战。对当前已有的点云法向量估计算法进行综述，分析其原理及关键技术，着重分析它们在处理噪声、外点和尖锐特征等方面的能力并给出比较，最后为未来研究提供了一些建议。

关键词: 三维点云, 法向量估计, 噪声, 外点, 尖锐特征