Inferência de orientação de dados esparsos para reconstrução de superfícies

Abstract

Orientation inference of sparse data for surface reconstructionThis work approaches the problem of sparse data spatial organization inference for surface reconstruction. We propose a variant of the voting method developed by Gideon Guy and extended by Mi-Suen Lee. Tensors to represent orientations and spatial infuence ¯elds are the main mathematical instruments. These methods have been associatedto perceptual grouping problems. However, we observe that their accumulation processes infer sparse data organization. From this point of view, we propose a new strategy for orientation inference focused on surfaces. In contrast with original ideas, we argue that a dedicated method may enhance this inference. The mathematical instruments are adapted to estimate normal vectors: the orientation tensor represents surfaces and infuence elds code elliptical trajectories. We also propose a new process for the initial orientation inference which effectively evaluates the sparse data organization. The presentation andcritique of Guy's and Lee's works and methodological development of this thesis are conducted by epistemological studies. Objects of different shapes are used in a qualitative evaluation of the method. Quantitative comparisons were prepared with error estimation from several reconstructions. Results show that the proposed method is more robust to noise and variable data density. A method to segment points structured on surfaces is also proposed. Comparative evaluations show a better performance of the proposed method in this application.