Hierarchical Structure Recovery of Point‐Sampled Surfaces

Abstract

We focus on the class of 'regular' models defined by Várady et al. for reverse engineering purposes. Given a 3D surface M represented through a dense set of points, we present a novel algorithm that converts M to a hierarchical representation H M . In H M , the surface is encoded through patches of various shape and size, which form a hierarchical atlas. If M belongs to the class of regular models, then H M captures the most significant features of M at all the levels of detail. In this case, we show that H M can be exploited to interactively select regions of interest on M and intuitively re-design the model. Furthermore, H M intrinsically encodes a hierarchy of useful 'segmentations' of M. We present a simple though efficient approach to extract and optimize such segmentations, and we show how they can be used to approximate the input point sets through idealized manifold meshes.

Fast hierarchical animated object decomposition using approximately invariant signature

“…Nevertheless, these two registration methods are not convenient for animated object decomposition because the correspondence optimization is sensitive to topological changes between the adjacent frames of animated object. Attene et al [2] proposed a hierarchical clustering for point-samples surfaces. However, this method only considers the single static point model and cannot be directly applied to the animated object.…”

“…Hence, the decomposition performance for the animated object is usually not efficient. (2) Unlike the video segmentation, where there exist pixel-to-pixel correspondences among the consequent image frames, the animated object sequences usually do not hold the point-to-point correspondence between the surfaces in adjacent frames. Especially for the point-sampled geometry sequences, the geometry objects in different frames usually contain different numbers of points, which increases the complexity of the decomposition.…”

Extraction of cylinders and estimation of their parameters from point clouds

“…Hierarchical clustering [12,1], which often requires that the number of segments be pre-selected by the user, is a good method for primitive extraction. However, this method is computationally intensive in time complexity and memory even if the model contains just a few thousand points, because the method investigates all types of primitives at the same time to create a global list of priorities by using algebraic fitting [13].…”

“…A cylinder is frequently described by 3 parameters: axis orientation a ! , a point on the axis p n and radius r [1]. However, few methods can determine these descriptive parameters accurately.…”

“…Most manufactured objects are composed of basic primitives such as planes, spheres, cylinders, toruses and cones [1,2]. Cylinders are the most frequently used primitive in these objects such as pipes, sleeves, connectors.…”