||Video stabilization as a variational problem and numerical solution with the Viterbi method
||Maurizio Pilu, PhD, Ing. Senior Research Scientist
||HP Labs Europe, Bristol, UK
||Video stabilization is the process of reducing
the amount of apparent image motion caused by
the secondary, unintentional camera motion with
respect to the scene while preserving the dominant,
intentional camera motion.
There are several situations where stabilization
based on classical filter theory (e.g. a dampened
integrator, time averaging, more general FIR/IIR
filtering, etc.) produces very good results for
reducing hand shake or in general when the range
of camera motion is not very big. However in cases
where the motion is dramatic, such as for wearable
video, there is a compromise to be made between
attenuating high frequencies and step response.
Besides there is always the problem of the constraints
imposed by the maximum size of the output video,
which are hard to enforce smoothly in classical filtering.
The work proposes a novel approach and a computational
method that produces the theoretically most stable
sequence allowed by the size of the output window
regardless of frequency of oscillations, type of
motion and extent of the image bounds. The method
is theoretically optimal in the situation where the
whole motion sequence is known in advance (off line
filtering) or sub optimal in the case of time-delayed
on-line filtering. A real time demo stabilization
demo will be shown.