Theorems and algorithms for multiple view geometry with applications to electron tomography
No Thumbnail Available
Doctoral thesis (monograph)
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Helsinki University of Technology Laboratory of Computational Engineering publications. Report B, 30
AbstractThe thesis considers both theory and algorithms for geometric computer vision. The framework of the work is built around the application of autonomous transmission electron microscope image registration. The theoretical part of the thesis first develops a consistent robust estimator that is evaluated in estimating two view geometry with both affine and projective camera models. The uncertainty of the fundamental matrix is similarly estimated robustly, and the previous observation whether the covariance matrix of the fundamental matrix contains disparity information of the scene is explained and its utilization in matching is discussed. For point tracking purposes, a reliable wavelet-based matching technique and two EM algorithms for the maximum likelihood affine reconstruction under missing data are proposed. The thesis additionally discusses identification of degeneracy as well as affine bundle adjustment. The application part of the thesis considers transmission electron microscope image registration, first with fiducial gold markers and thereafter without markers. Both methods utilize the techniques proposed in the theoretical part of the thesis and, in addition, a graph matching method is proposed for matching gold markers. Conversely, alignment without markers is disposed by tracking interest points of the intensity surface of the images. At the present level of development, the former method is more accurate but the latter is appropriate for situations where fiducial markers cannot be used. Perhaps the most significant result of the thesis is the proposed robust estimator because of consistence proof and its many application areas, which are not limited to the computer vision field. The other algorithms could be found useful in multiple view applications in computer vision that have to deal with uncertainty, matching, tracking, and reconstruction. From the viewpoint of image registration, the thesis further achieved its aims since two accurate image alignment methods are suggested for obtaining the most exact reconstructions in electron tomography.
robust regression, robust estimation, statistical modeling, epipolar geometry, fundamental matrix, uncertainty, image matching, affine reconstruction, affine triangulation, degeneracy, bundle adjustment, image registration, image alignment, computer vision, electron tomography