Estimación de normales y reducción de datos atípicos en nubes de puntos tridimensionales

Abstract

This paper shows a method for normal vectors estimation and outlier reduction for point cloud surfaces. This method is based on Robust Principal Component Analysis applying to a sizable neighborhood, estimated according to local geometric featuring. First, outlier reduction is applied by means of a Mahalanobis distance based procedure. Subsequently, a variable size neighborhood is estimated calculating local geometric features. The resizable neighborhood showed to give a more accurate tangent plane estimation than a size fixed neighborhood, allowing accurate normal estimation. The method permits normal vector estimation that is robust to noise, outliers and border and sharp features.