Abstract
Because of their relative ease in solving the correspondence
problem, stereo systems without a relative rotation are popular. However,
in practice, mechanical difficulties will lead to a small, unknown
relative rotation between stereo cameras. In this paper we present an
algorithm for the calibration of a stereo system with small relative
angles in an uncontrolled environment. This algorithm has two advantages:
(a) It is more accurate than the existing algorithms in the computer
vision and photogrammetry literatures. (b) It provides useful insight into
the problem of camera calibration and relative orientation. This is done
by deriving explicit analytical solutions for the relative pan, tilt, and
roll angles in terms of the world pan angle (gaze angle) and the
coordinates of the feature points used in their computations. These
solutions allow us a better understanding of the problem of calibration in
general by providing us with the insight as to how errors due to
quantization and uncertainty in the location of image centers affect the
computation of rotation angles. It is shown that as the distance features
points from the center of the image decreases, the error due to
quantization in the relative pan angle increases quadratically, that of
the relative roll angle increases linearly, while that of the tilt angle
does not change appreciably. Likewise, it is shown that errors in the
locations of principal points (image centers) do not affect the
computation of relative pan and roll angles appreciably, whereas the
impact on the relative tilt angles is significant. These findings are
likely to be of use even when the relative rotation angles are not small.
All of the analytical findings have been supported by extensive
simulation.
Intelligent Decision Aids
Intelligent M4 Systems
Machine Learning
Neural Networks
Back to NCARAI Library