特征敏感的点云重采样算法

Point Cloud Resampling Algorithm of Feature-sensitivity

针对原始点云数据采样不均匀,数据缺失等问题,提出一种基于特征敏感的点云重采样算法.该算法首先运用主成分分析法获得点云的初始法向量,将高斯映射和空间密度权重相结合,自适应地将点云划分为特征点和非特征点;其次,采用各向异性的相邻点优化特征点的拟合平面,提高特征点的法向量准确度;在此基础上,通过带空间权重的投影插值算法和在加权局部最优投影(WLOP)算法中引入法向权重,实现了特征保持的点云均匀采样.实验结果表明:与经典WLOP算法相比,该算法在均匀采样的同时,能够以较高的压缩率对点云向下采样并保持点云特征,向上采样时可以对缺失点云进行有效修复,有利于点云数据的后续处理.

The raw scanned point clouds of objects are often characterized with non-uniform point density and data lost. A point cloud resampling algorithm based on feature-sensitive is proposed. First, principal component analysis method was used to estimate normal for points, the raw point cloud was divided into feature points and non-feature point by Gauss map with adaptively spatial density weights. Then, the fitting plane of feature points was iteratively optimized by anisotropic neighborhood points, the normal can be accurately estimated by the optimization fitting plane. On this basis ,using the progressive projection algorithm with density weights upsample the feature points, and normal difference weights was incorporate into weighted Locally Optimal Projection operator, the algorithm produced clean, uniform and feature-preserved point cloud. Compared with the classical WLOP algorithm, the experimental results show that the algorithm can downsample the point cloud in high compress ratio with feature-persevered, and it can interpolate the miss- ing points while upsampling procedure, the point cloud post-processing would benefit from the uniform and feature-preserved point cloud.