This thesis presents a statistical learning framework for inferring geometric structures from images. Specifically, the proposed framework computes dense range maps of location sin the environment using only intensity images and very limited amount of range data as an input. This is achieved by integrating and analyzing the statistical relationships between the visual data and the available depth on terms of small patches. The scientific issue is to represent this correlation such that it can be used to recover range data where missing. Markov Random Fields are used as a model to capture the local statistics of the intensity and range. / Experiments on real-world data are conducted under different configurations to demonstrate the feasibility of the method. In particular, our application is in mobile robotics, where inferring the 3D layout of indoor environments is a critical problem for achieving exploration and navigation tasks. The modeling of a large-scale environment involves the acquisition of a huge amount of range data to extract the geometry of the scene, and is often performed using sophisticated but costly hardware solutions. This task is physically demanding and time consuming for many real systems. By using the proposed framework, it is demonstrated that we can learn the geometric characteristics of the environment from the incomplete sensory data to build a 3D model of it. / The contributions of this thesis are mainly three: First, it demonstrates the viability of the use of very limited range data together with intensity to recover complete dense range maps. Second, it presents a complete framework for building a 3D model of an indoor environment using a mobile robot. And third, it analyses and outlines the advantages and limitations encountered when dealing with large indoor environments. / An additional contribution is the use of the method we propose for range estimation to an alternative problem: color correction and augmentation with the specific application to underwater images.
| Identifer | oai:union.ndltd.org:LACETR/oai:collectionscanada.gc.ca:QMM.102218 |
| Date | January 2005 |
| Creators | Torres Méndez, Luz Abril. |
| Publisher | McGill University |
| Source Sets | Library and Archives Canada ETDs Repository / Centre d'archives des thèses électroniques de Bibliothèque et Archives Canada |
| Language | English |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Format | application/pdf |
| Coverage | Doctor of Philosophy (School of Computer Science.) |
| Rights | © Luz Abril Torres Méndez, 2005 |
| Relation | alephsysno: 002339154, proquestno: AAINR25269, Theses scanned by UMI/ProQuest. |
Page generated in 0.0021 seconds