UvA-DARE is a service provided by the library of the University of Amsterdam (https://dare.uva.nl)
UvA-DARE (Digital Academic Repository)
Intrinsic statistical techniques for robust pose estimation
Dubbelman, G.
Publication date
2011
Link to publication
Citation for published version (APA):
Dubbelman, G. (2011). Intrinsic statistical techniques for robust pose estimation.
General rights
It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).
Disclaimer/Complaints regulations
If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible.
Summary
The context of this Ph.D. thesis is that of intelligent mobile systems. For many of their applications determining the geometrical structure of their environment and their pose is crucial. In this thesis we focus on computer vision based methodologies to estimate the pose of a moving camera from its recorded image data. Fundamental principles which limit the accuracy, stability and efficiency of current state-of-art solutions are presented. We show that more accurate and efficient methods can be developed by exploiting the mathematical structure of pose spaces within statistical computations. Three such novel algorithms are presented in this thesis. The relevance of their obtained improvements is evaluated on challenging simulated data and on real image data. The latter includes image sequences recorded by a camera mounted on a vehicle which drove a total of 7.5 km. Be-sides experimental validations, our methods are also provided with extensive theoretical foundations. The scientific contributions contained within this Ph.D. thesis are summa-rized below.
1. A novel distance measure is presented for the epipolar manifold. It is based on a multiple geodesic approach and facilitates ML solutions. A previously proposed distance measure for the epipolar manifold is proven mathematically incorrect. It is shown experimentally that this incorrect distance measure does not facilitate ML solutions.
2. A novel verification free pose estimator is developed. It is specially designed to per-form visual-odometry. Instead of finding consensus in image space, as RANSAC does, this approach finds consensus in pose space. This is significantly more ef-ficient as it does not require hypotheses verification. Its accuracy is shown to be favorable over that of various RANSAC approaches and that of non-linear mean-shift.
3. A novel algorithm is proposed to enhance accuracy and stability of state-of-the-art RANSAC approaches which estimate the epipolar geometry. It is facilitated by our novel distance measure over the epipolar manifold. Its accuracy is favorable over that of various RANSAC approaches, including those using local iterations, and that of non-linear mean-shift.
4. A novel closed-form trajectory bending algorithm is presented. It can perform loop-closure and sensor-fusion with linear complexity in the number of poses. It is proven that the cost function minimized by our algorithm is a simplification of that of an existing maximum likelihood approach. It is shown that under realistic conditions our closed-form method provides similar and sometimes even favorable accuracy. It is also significantly more efficient.
The improved accuracy and efficiency obtained by our methods bring closer the actual application of intelligent mobile systems.
