Abstract: Aiming at the conflicts between smoothing of low frequency area and feature preserving of high frequency area of classical point cloud denoising algorithms, a new algorithm is proposed based on robust regression of local adaptive neighborhood. The concept of local adaptive neighborhood of a sample point is proposed, the range of the neighborhood can be ajusted adaptive by the shape of local area, providing bases for smoothing of low frequency area and feature preserving of high frequency area. To solve the problem of easy influenced by outliers of the least squared algorithm, leading to less robustness of differential geometric information estimation，the robust regression algorithm is adopted, improving the reliability of estimated result. A new feature measure is proposed based on norm, min-max curvature of the sample point. The characteristics of the measure are discussed, and the sample points are classifed into feature points, non-feature points and transitional points. The effective neighborhood of the points is recognized based on feature measure and smoothing of low frequency area and feature preserving of high frequency area are realized. The effectiveness of our method is validated by comparison experiments.

Key words: Feature measure, Local adaptive neighborhoods, Point clouds, Robust regression