Filtering methods of scattered point set based on multi-scale kernel function
基于多尺度核函数的散乱点云数据过滤方法*

This paper proposed a method based on multi-scale kernel function for robust filtering of a noisy set of points sampled from a smooth surface. The method used a kernel density estimation technique and a mean-shift algorithm for point clustering. With every point of the input data, associated a local likelihood measure capturing the probability that a 3D point was located on the sampled surface. The remaining set of maximum likelihood points deliverd an accurate point-based approximation of the surface. Some established meshing techniques work well in conjunction with the filtering method for surface reconstruction. Experiment results show that the filtering procedure suppresses noise of different amplitudes and allows for an easy detection of outliers which are then automatically removed.