National Technical University of Athens (NTUA)
Γεωμετρική πληροφορία από μεμονωμένες εικόνες στην φωτογραμμετρία και την όραση υπολογιστών
Abstract
dc:descriptionMotivation of this thesis is to investigate the geometry of single perspective images in three main aspects. First, the aim is to further understand this geometry by formulating alternative approaches for certain significant problems, regarding particularly the geometry of the image vanishing points and their relation with the question of camera calibration, on the one hand, and with issues of object reconstruction, on the other. A second important aspect was to carry out this research in the typically Euclidean conceptual framework of Photogrammetry, but at the same time to clearly interpret the relation of the algorithms developed here with corresponding approaches from the field of geometric (computational) Computer Vision, where single images are mainly studied. In this sense, the thesis at the same time presents Euclidean interpretations, useful to Photogrammetry, of methods Computer Vision which, as a rule, rely on projective geometry. Finally, of course, an important goal of this thesis was to implement the theoretical investigation in the form of specific automatic algorithms, in order to thereby demonstrate its direct practical usefulness. Thus, the thesis starts with certain introductory comments regarding similarities and differences between Photogrammetry and Computer Vision, as well as the theoretical and practical significance of projective geometry, while certain fundamental properties of projective geometry, necessary for the following, are also presented. Next, the geometry of an image with three vanishing points of orthogonal space directions and its relation to intrinsic camera geometry as well as to homography are studied, while different algorithms for camera calibration via vanishing points - one of which allows simultaneous estimation of radial-symmetric lens distortion - are evaluated against self-calibrating multi-image bundle adjustment. After this, the geometric properties and the equations which relate camera interior orientation with the vanishing points of two orthogonal directions are formulated. This relation is expressed by the introduction of a geometric locus for the projection centre (“calibration sphere”), which is compared to corresponding projective representations from Computer Vision. On this basis a camera calibration algorithm is formulated and evaluated experimentally, which allows combining different single images with two vanishing points from the same camera. Then the algorithm is generalized for automatic camera calibration using one image with three vanishing points of orthogonal directions or more independent images with at least one pair of vanishing points. The approach is fully automated with the automatic localization of vanishing points on one or more images via a novel technique and a unified least-squares adjustment of observations on all participating images through the line-fitting and “calibration sphere” equations. Results from such automatic calibrations using real data were regarded as satisfactory compared to those from multi-images self-calibration as regards both calibration as well as 3D object reconstruction. The thesis addresses next the question of affine reconstruction of planar objects. The geometry and equations are explained which allow affine reconstruction of uncalibrated images, both frontal and oblique with different orientation, through one single vanishing point. The algorithm (with exclusive metric information a dimension in the direction of the road) is applied and evaluated in cases of vehicle speed measurement from video sequences. In this algorithm moving vehicles are detected and tracked automatically on affine image rectifications, while a shadow detection technique is also incorporated. In this way, it is demonstrated that purely projective approaches have also a direct practical relevance for Photogrammetry. The thesis closes with the conclusions from this research regarding the theoretical and practical significance of single images in different photogrammetric tasks. Finally, certain interesting questions for further investigation are indicated, particularly regarding accuracy issues of the approaches.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2007
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Grammatikopoulos, Lazaros
- Γραμματικόπουλος, Λάζαρος
Subjects
dc:subject × 20- Φωτογραμμετρία
- Όραση υπολογιστών
- Βαθμονόμηση μηχανής
- Σημεία φυγής
- Ευθεία φυγής
- Σφαίρα βαθμονόμησης
- Προβολική γεωμετρία
- Αφινική ανακατασκευή επίπεδων αντικειμένων
- Photogrammetry
- Computer vision
- Single image calibration
- Vanishing points
- Vanishing line
- Calibration sphere
- Projective geometry
- Afine reconstruction of planes
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Πολιτικού Μηχανικού
- Engineering and Technology
- Civil Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/16955