Título: Three-dimensional model construction from multiple sensor viewpoints
Autores: Aubry, Stéphane
Fecha: 1994
Publicador: McGill University - MCGILL
Fuente:
Tipo: Electronic Thesis or Dissertation
Tema: Engineering, Electronics and Electrical.
Descripción: This work addresses the automatic construction of geometric models of real scenes, from multi-view three-dimensional sensor data.
We review range data acquisition, multi-view integration, and solid modeling. We show that knowledge about the data acquisition procedure yields not only the coordinates of the acquired points, but also additional geometrical information. We use that information to draw a geodesic proximity graph with respect to the surface the data points lie on. Such a graph is useful because it provides the connectivity information necessary for subsequent differential-geometric processing, and three-dimensional surface modeling. We say that the graph is a least-commitment boundary representation, because it does not involve the use of higher level or cognitive processes.
We specialize the concept to the common case of line-of-sight optical sensors. We give formal definitions of graph validity using assumptions of object opacity and object rigidity, and we demonstrate that Euclidean proximity graphs drawn on the data points are not valid when the data is sparse with respect to surface concavities.
We describe a sub-quadratic incremental view integration algorithm that assumes the data is highly-organized. It guarantees graph validity under restrictive conditions. Therefore, we present another non-incremental algorithm with no assumption on the input data organization. It is based on an iterative carving of the graph faces, starting with the convex hull of all points as the initial model. It builds a hierarchy of models, each of which is internal to the previous one. We test the algorithm with real data on an object homeomorphic to a sphere, and incorporate heuristics designed to yield more geometrically pleasing results.
Idioma: en