We present a very simple and effective method for eliminating the degeneracy inherent in a planar scene, and demonstrate its performance in a useful application -- that of binocular self-calibration. The projective geometry of planar scenes suffers from a two-fold ambiguity: projective structure cannot be recovered from a pair of images of the scene, and transformations in projective 3-space are underconstrained for features lying on a plane. We overcome both these problems by combining feature data from different pairs of images where we know the relationship between the images to be identical. In essence this generates images of a non-degenerate scene which can then be processed using standard algorithms. The only constraint is that the binocular head must be able to make repeated identical rotations about its axes. The benefits of this technique are demonstrated by its use in solving the problem of self-calibration. The use of data combination enables standard general scene stereo ...
Joss Knight, Ian D. Reid