Back to search

National Technical University of Athens (NTUA)

Η επιπολική γεωμετρία στον προβολικό και τον ευκλείδειο χώρο

Abstract

dc:description

Subject of this PhD Thesis is the geometry of the uncalibrated stereopair. In particular, the question is addressed as to which information may be extracted from two overlapping images regarding their position in space, the cameras from which they originate and the recorded objects. Starting point was the treatment of this issue in projective space both (today) by researchers in the field of computer vision as well as (more than a century ago) by pioneering photogrammetrists; but, a basic aim here was to simultaneously handle the question also in the Euclidian space. In the Introduction, the subject of the thesis is posed, its wider research framework is defined through a review of related literature in the fields of photogrammetry and particularly computer vision, and the original aspects of this work are outlined. Next, certain introductory concepts of projective geometry, indispensable for the description and further study of image geometry are given. Following this, the epipolar geometry of the stereopair is presented in the projective framework of computer vision. For calibrated images the linear expression of the coplanarity condition and relative orientation represented by the essential matrix is described and its properties are given. Also, the concept of 2D epipolar geometry for uncalibrated camera pairs is introduced. This is expressed by the fundamental matrix which describes, directly on the two image planes, the one-way correspondence between image points and epipolar lines. Different geometric interpretations of the fundamental matrix are presented along with algorithms for its computation from point correspondences as well as for projective reconstruction of the imaged space but also for its upgrading to affine or Euclidean through geometric constraints referring to object space or the camera parameters. Further, constraints are given between the fundamental matrix and interior orientation, which allow camera autocalibation from ≥3 images or partial camera calibration in the case of the stereopair. The following chapters present the novel contributions of this thesis. Given the two epipoles and the projective correspondence between homologue epipolar lines - information inscribed in the fundamental matrix - the 2D epipolar geometry is studied in the Euclidean 3D space. Specifically, a mathematical expression is formulated to encompass the infinite, compatible with the fundamental matrix, combinations of relative and interior orientation of the two images and, consequently, the corresponding projective reconstruction of imaged space. This description relies on the definition of 4 geometric parameters of the stereopair (direction of the intersection line of the image planes; angle formed by these planes; location of projective centres on the base-line) which are independent of the fundamental matrix. Furthermore, geometric loci are established which describe changes of these parameters but also different constraints on camera geometry parameters. In this context, an alternative geometric parameterization of the 7 degrees of freedom of the fundamental matrix is formulated, based on the proof that, with suitable 2D translation and rotation, two planar projective bundles of rays - which generally intersect on a conic section - may be brought to a given perspective position, i.e. to intersect on a straight line. On the other hand, with a different but still Euclidean approach equations are formulated to express in a new form the constraints posed by the fundamental matrix upon the parameters of interior orientation. These equations rely on the equality between angles formed by the faces of tetrahedra which are defined, independently for the two images, by specific epipolar planes and the planes formed by the optical axes and the base of the stereopair. Based on these equations, four new closed-form algorithms for partial camera calibration are formulated. In the last part, the algorithms developed in this thesis are evaluated with both simulated as well as real image data (including images from datasets available on the Internet). The computation of the fundamental matrix from automatically extracted image point correspondences is examined; examples of image configurations in 3D Euclidean space compatible with specific 2D epipolar geometries are given; the algorithms for partial camera calibration are compared with those from computer vision literature. The thesis is completed with conclusions from this study, justification of the originality of the performed research work and certain thoughts regarding further possible research issues in the context of the present investigation.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2010

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Kalisperakis, Ilias
  • Καλησπεράκης, Ηλίας

Subjects

dc:subject × 18

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/21678

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Kalisperakis, Ilias; Καλησπεράκης, Ηλίας. Η επιπολική γεωμετρία στον προβολικό και τον ευκλείδειο χώρο. National Technical University of Athens (NTUA), 2010. http://hdl.handle.net/10442/hedi/21678