Veselin Dikov
Veselin Dikov did his Master's thesis under supervision of Prof. Navab (Chair for Computer Aided Medical Procedures & Augmented Reality) in cooperation with MVTec Software GmbH. He graduated from CSE/BGCE in November, 2006, and is now Software Engineer at MVTec.
Master's Thesis:
Stereo Vision with Conical and Circular Features
Stereo vision is a major subject in the scientific field of computer vision. The main task in stereo vision is the 3D reconstruction of a scene from given images taken by two cameras at different locations. First of all, however, the geometrical relations between cameras, represented by the so-called epipolar geometry, need to be found. Traditional algorithms that estimate the epipolar geometry between two images are based on point image features.
A stereo vision scene with elliptical features
Using ellipses and circles as image features, instead, has certain advantages compared to using point features: ellipses and circles can be extracted more easily and with higher accuracy from the image, even when they are partially occluded. Moreover, they can be easier identified in multiple images. Such features can be found in many man-made objects, even if point features are completely missing. Finally, it is theoretically proven that only four conical features in general position are sufficient to estimate the epipolar geometry; however, unfortunately, not in a linear manner.
Estimation of epipolar geometry with plane and parallaxapproach (three ellipses whould be sufficient)
New ideas for the linear handling of this estimation problem, at least for specific cases, are developed in this thesis. An algorithm is proposed that finds groups of ellipses that lie on a common plane in space. Based on the results of this algorithm, linear methods for \ estimating the epipolar geometry are developed. The thesis gives a proof that a minimal configuration of three ellipses is sufficient, provided that two of the ellipses lie in the same plane. For this puropse, self-calibrations method are deployed, which were not used in combination with binocular stereo vision so far. If the cameras are known (esp. their focal length and their relative position), then two coplanar ellipses or two circles in general position are also sufficient.
