MODELS AND METHODS FOR GEOMETRIC COMPUTER VISION, ACTA UNIVERSITATIS OULUENSIS C Tec h n i c a 3 5 3

Automatic three-dimensional scene reconstruction from multiple images is a central problem in\ngeometric computer vision. This thesis considers topics that are related to this problem area. New\nmodels and methods are presented for various tasks in such specific domains as camera\ncalibration, image-based modeling and image matching. In particular, the main themes of the\nthesis are geometric camera calibration and quasi-dense image matching. In addition, a topic\nrelated to the estimation of two-view geometric relations is studied, namely, the computation of a\nplanar homography from corresponding conics. Further, as an example of a reconstruction system,\na structure-from-motion approach is presented for modeling sewer pipes from video sequences.\nIn geometric camera calibration, the thesis concentrates on central cameras. A generic camera\nmodel and a plane-based camera calibration method are presented. The experiments with various\nreal cameras show that the proposed calibration approach is applicable for conventional\nperspective cameras as well as for many omnidirectional cameras, such as fish-eye lens cameras.\nIn addition, a method is presented for the self-calibration of radially symmetric central cameras\nfrom two-view point correspondences.\nIn image matching, the thesis proposes a method for obtaining quasi-dense pixel matches\nbetween two wide baseline images. The method extends the match propagation algorithm to the\nwide baseline setting by using an affine model for the local geometric transformations between the\nimages. Further, two adaptive propagation strategies are presented, where local texture properties\nare used for adjusting the local transformation estimates during the propagation. These extensions\nmake the quasi-dense approach applicable for both rigid and non-rigid wide baseline matching.\nIn this thesis, quasi-dense matching is additionally applied for piecewise image registration\nproblems which are encountered in specific object recognition and motion segmentation. The\nproposed object recognition approach is based on grouping the quasi-dense matches between the\nmodel and test images into geometrically consistent groups, which are supposed to represent\nindividual objects, whereafter the number and quality of grouped matches are used as recognition\ncriteria. Finally, the proposed approach for dense two-view motion segmentation is built on a\nlayer-based segmentation framework which utilizes grouped quasi-dense matches for initializing\nthe motion layers, and is applicable under wide baseline conditions.

ISBN-10:
978-951-42-6150-3
Kieli:
eng.
Sivumäärä:
190 s.
Tekijät:
Kannala Juho
Tuotekoodi 013962
20,00 €