Robust Method for Estimating Normals on Point Clouds Using Adaptive Neighborhood Size

Abstract

Normal estimation on sampled curves or surfaces is a basic step of many algorithms in computer graphics, computer vision, and especially in recognition and reconstruction of three dimensional objects. This paper presents a simple and intuitive method for estimating normals on point based surfaces. The method is based on Robust Principal Component Analysis (RPCA) therefore is capable to deal with noisy data and outliers. In order to estimate an accurate normal on a point, our method takes a neighborhood of variable size around the point. The neighborhood size depends on local properties of the sampled surface. It is shown that the estimation of the tangent plane on a point is more accurate using a neighborhood of variable size than using a fixed one.