散乱点云去噪算法的研究与实现

提出了一种快速去除散乱点云数据表面噪声和离群点的鲁棒滤波算法.应用核密度估计聚类方法,通过Mean-Shift迭代过程将每一个采样点"漂移"到核密度估计函数的局部最大值点,该最大值点确定了点云数据的聚类中心并能准确逼近原始曲面,使点云曲面收敛为一个稳定的三维数字模型.算法中的似然估计函数充分考虑了散乱点的法矢方向,因此不仅可以去除不同幅度的噪点,还可以用简单的阈值条件很容易地检测出离群点的聚类,从而实现了点云数据的高效快速光顺去噪.

Research and implementation for denoising noisy scattered point data

A method for robust filtering of a noisy set of points sampled from a smooth surface is presented.A kernel density estimation technique is used for point clustering in the presented method.Each sample point is shifted to the local maximum of the kernel function by a mean-shift based clustering procedure.The clustering center of point cloud is confirmed through the remaining set of maximum likelihood points,and the point-based surface is also approximated accurately by the same way,so the point-set surface can be converged to a stable 3D digital model.The normal directions estimated at the scattered points are concerned in the likelihood function,so noise with different amplitudes can be suppressed during the filtering procedure.Outliers can be easily detected and automatically removed by a simple threshold in the algorithm,so robust filtering of noisy scattered point cloud data is implemented.