Abstract:

This paper presented a feature-preserving algorithm to smooth 3D point cloud model. It began with constructing kd tree(k-dimensional tree) of the model, then computed the k-nearest neighbor of sampling point. Local geometry information was defined as the feature parameters, which considered the average distance, the normal angle among the point and its neighboring points, point curvature parameters. Then the sampling point was moved according to its characteristics and normal direction. Experimental results show that this method is effective to remove the noise point and preserve the feature of the original point cloud model.

Key words: point cloud smoothing, curvature, principal component analysis, feature-preserving

摘要：

提出了一种特征保持的三维点云光顺去噪算法。首先对点云模型构造kd树结构,计算采样点的k邻域,然后把点云模型的局部几何信息,包括邻域点间距离、法向夹角、曲率等作为特征参数,根据其特征性强弱将采样点沿法向方向移动不同距离来实现点云去噪。实验结果表明,算法既能有效去除噪声,又能很好地保留原始模型的特征信息。

关键词: 点云光顺, 曲率, 主成分分析, 特征保持