Calibrated Zoom for Stereo Vision
Brent Seales
University of Kentucky
2:30pm Friday March 24 in room 2310
Active camera control can improve the quality of the images being used
for visual tasks. For example, zooming can improve the resolution of
an image feature, allowing for highly-localized stereo matches.
In this talk I will present a new approach to the problem of calibrating
a zooming camera for the purpose of stereo vision. This approach is
different from other methods in two primary ways. First, we avoid placing
a heavy dependence on individual, strongly localized features. Feature
localization is obtained iteratively, formulated as part of an error
criterion used in various passes of an optimization process. Second, this
staged optimization process considers all images simultaneously, representing
the parameters of the final calibrated camera as a function of zoom.
Experiments on a real image sequence taken by an active, zooming stereo rig
demonstrate that with this method one can localize epipolar (stereo)
correspondence at any zoom setting to an accuracy of 1 pixel.
At the end of the presentation I will also outline a simple approach to
the time-to-contact problem using active zoom control. Our experiments
show that using the active zoom gives better accuracy over a longer depth
range than fixed-focal-length lenses.