基于APOCS的点云去噪算法

Denoising of point cloud based on Average Projections Onto Convex Sets

使用3D扫描工具获取点云数据时常常会伴随着加性噪声。本文提出了一个基于凸集平均映射(APOCS)的原创点云去噪算法。首先,使用基于相关性筛选改良过的均值滤波算法对噪声点云的法向量进行滤波。每个滤波后的法向量都能决定其相对应的局部平面,将这些平面视为凸集,可通过APOCS算法更新点云中各点的位置。在本文的最后将给出该算法在不同强度噪声点云上与双边滤波对比的实验结果。

Point cloud data is often accompanied by additive noise when it is collected by 3 D scanning tools. This paper proposes an original method of point cloud denoising based on APOCS（ Average Projections Onto Convex Sets）. First, the normal vectors of the noised point cloud is filtered using a mean filter improved by correlation based selection. And some planes are determined by the filtered normal vectors,which can be used to optimize the points ＇ locations through APOCS. Compared with bilateral filtering,the experiment demonstrates the results on point clouds with different intensity noises.