|
Identifier |
uch.csd.msc//1999tzurbakis |
Title |
Αντιστοίχιση Χαρακτηριστικών με Χρήση Γεωμετρικών Περιορισμών σε Εικόνες που Περιλαμβάνουν Δύο Τουλάχιστον Επίπεδα |
Alternative Title |
Feature Correspondence Using the Geometry of Scenes Containing at least Two Planes |
Creator |
Tzurbakis, Stavros V
|
Abstract |
In this thesis we address the problem of establishing feature correspondences between different views of a 3D scene. It is well known that this is one of the fundamental problems in computer vision since a lot of applications like camera callibration, recognition, scene reconstruction, view synthesis and others, depend on its solution. Existing methods exploit photometric constaints, the geometry of the camera or the geometry of the scene to come up with a solution to this problem. The major disadvantage in most of such methods is that their undelying assumptions are valid only in the case of images that are acquired from similar viewpoints (e.g. images that are acquired by a stereoscopic configuration). We propose two new methods for the problem of establishing point correspondences between different views. The first method exploits the geometry of the stereoscopic camera configuration while the second exploits constraints imposed by the geometry of the imaged scene. Both methods belong to the general class of uncalibrated vision techniques and their main advantage is that they can be used in scenes where the viewing conditions have been changed significantly dewring their capture. The methods were tested on a variety of indoor scenes and their performance was compared with the performance of two existing techniques. The experimental results show the effectiveness of the proposed methods. Finally, we discuss potential applications of the proposed methods and we propose extensions of this work.
|
Issue date |
1999-11-01 |
Date available |
1999-11-24 |
Collection
|
School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses
|
|
Type of Work--Post-graduate theses
|
Views |
517 |