Ing. Čeněk Albl presents 3D reconstruction with time variant geometry

On 2019-08-19 14:00:00 at E112, Karlovo náměstí 13, Praha 2
Methods for reconstructing 3D models of our environment have been long
established and well studied. The majority of 3D computer vision techniques
rely on the perspective camera model, which does not change its parameters over
time. However, there are many cases where this assumption does not accurately
fit the reality of the image capture process. For example, a majority of
today's cameras use CMOS sensors with a rolling shutter, which exposes every
image line at a different time. When such a camera is moving, e.g., in
hand-held videos,aerial footage, or augmented reality applications, the images
will contain distortions.These distortions cause standard computer vision
algorithms to fail. We need a different camera model that can describe the
camera motion as well in order to make the standard methods work to our
satisfaction. This work proposes new rolling-shutter adapted algorithms and
presents models that describe camera motion during the image capture and,
therefore, provide improved results.

Supervisor: doc. Ing. Tomáš Pajdla, Ph.D.
Supervisor - specialist: RNDr. Zuzana Kúkelová, Ph.D.
Responsible person: Petr Pošík