Hierarchical PCA decomposition of point clouds

Abstract

We present a hierarchical analysis technique for point clouds, based on Principal Component Analysis (PCA), a well known multivariate statistical method. The crux of the algorithm is a top-down planarity assessment of the underlying point data, after which individual planar patches are merged using a tree clustering technique. We will demonstrate how the results of this analysis are used as a preprocessing step for computer aided inspection of sheet metal folding, surface reconstruction and a hybrid point-polygon rendering algorithm.