基于曲率特征混合分类的高密度点云去噪方法

为解决复杂曲面点云在平滑去噪中存在的问题，提出基于曲率信息混合分类的特征保持点云平滑算法．该方法将平面投影与双边滤波算法相结合，采用主成分分析法对点云的局部曲率特性进行评价，使用线性组合混合分类方法将数据分为平面、次特征、富特征类型以及组合类型．针对不同特征邻域类型，提出平面类型的投影平滑方法、次特征和富特征类型的变参数双边滤波法平滑方法的线性组合方法实现点云数据的平滑去噪．将该方法用于激光三维高分辨率人体扫描系统所得到的高密度点云数据，实验结果表明该方法能够在有效光顺点云的同时保持其表面的几何特征，且简化了法向调整的繁杂运算．

Denoising Approach of High Mixed Classification Density Point Clouds Based on of Curvature Features

In order to resolve the existing problems in algorithms for smoothing and denoising complex point clouds, an effective and feature-preserving 3D point-cloud smoothing method based on the mixed curvature feature classification was presented. It combined local plane projection and bilateral denoising method. The local curvature feature of each point was evaluated by principal component analysis. Ac- cording to the features, the point clouds were divided into plane, sub-featured and rich-featured classes and linearly mixed type. With the different neighborhood types, the combination of the plane projection smoothing method for plane class and the variable parameter bilateral filtering method for sub-featured and rich-featured classes were applied. The algorithm presented has been applied to process point-cloud data sets obtained by a high resolution 3D human body laser scanner. The experimental results show that the proposed method can effectively remove noise and preserve geometric features while can significantly sim- plify the normal direction adjustment process.