DOI: https://doi.org/10.15368/theses.2008.11
Available at: https://digitalcommons.calpoly.edu/theses/13
Date of Award
6-2008
Degree Name
MS in Computer Science
Department/Program
Computer Science
Advisor
Zoë Wood
Abstract
Laser range scanning is one of the leading methods for the acquisition of 3D models from real world objects. This process, however, introduces significant excess topological handles which increases the complexity of future processing, and lowers the quality of the acquired models. Previous research has shown that the hole filling step of the model creation pipeline is the primary cause of excess handles. We explore the hole filling process in detail and discuss the limits of hole fillers that work on the reconstructed surface and of those that work in the volumetric setting. In addition, we present our algorithm which aims to reduce the excess handles by adapting and improving filters that work in the volumetric domain to fill holes in the scanned data. Using these filters we are able to reduce the topological noise by 47% and to improve the output appearance of surfaces processed by existing hole fillers.