Abstract: Concerning the point clouds with noises and outliers acquired by a 3D scanner, a denoising method based on the concept of local outlier was proposed. The method established the topology connection of the scattered points by searching the k-Nearest Neighbor (kNN) of each point. Local outlier factor was calculated to weight the current point's outlier level, so the noises can be restricted and the outliers can be removed. The method emphasized how to detect the outliers and noises with low density when they are scattered around the point cloud with high density. The experimental results show that the method can detect the outliers next to model boundaries easily, and maintain the borders to the greatest extent.

Key words: local outlier, k-Nearest Neighbor (kNN), model boundary, denoising

摘要： 针对三维扫描仪获取的带噪声和离群点的点云数据，提出了基于局部离群点概念的去噪算法。通过k-近邻(KNN)搜索建立散乱点之间的拓扑关系，进而计算当前测点的局部离群因子以衡量该点的离群程度，从而限制噪声并剔除离群点。重点解决了高密度扫描点云周围分布的低密度离群噪声点的识别问题。实验结果证明，该算法能有效检测出紧挨模型边界的噪声点，并最大限度地保持模型边界。

关键词: 局部离群点, k-近邻, 模型边界, 去噪