Abstract:
Using free-formed surface mirrors into optical systems introduces
difficulties. Hence, so far such reflections were only regarded as
source of error in image processing. Accordingly, there are no methods
to extract the information contained in the reflection image. The
problems and methods are analyzed in the first part of this
thesis. The camera calibration and the surface reconstruction, the
basic requirements for every reconstruction methods, form the second
part of the thesis. A new method for the imaged-based data-acquisition
in the calibration effort is presented. It improves the existing
calibration technique manifold. On the basis of this technique, a new
method to recover a reflective surface's topography is elaborated.
The last part is devoted to the detailed presentation of two methods
for the extraction of the contained information. The first method is a
non-classical approach in the sense that the distorted image is not
reconstructed to a pinhole-camera view. It depends on an extension of
a usually applied linear constraint in stereo-image processing to a
more general non-linear understanding of the restriction. The second
one is a classical reconstruction of the image, based on the
comparison of the reflection directions of the free-formed surface and
a virtual planar mirror. This method is implemented in two ways. One
is an ordinary implementation on the central processing unit of the
computer. The second one uses the graphics processing unit (GPU),
i.e. the main computational unit on the computer's graphics-card, to
reconstruct the image. The usage of the GPU offers manifold new
possibilities for image processing in general. Both implementations
are tested on images acquired in the experimental vehicle. For one
method a coarse distance determination is conducted. This thesis
proves that reflections on free-formed surface mirrors are sources of
information, even if the free-formed surface was not constructed to
serve as a mirror in an optical system.