We propose a new approach to the fundamental matrix estimation problem in s
tereo vision, which is based on the optimal weighting of the matching point
s by their a posteriori probability to be relevant. With the affine camera
model the method produces better results compared to the known robust metho
ds including those with the full camera model. The same approach with the f
ull perspective projection case is much more difficult to solve, however, t
he results obtained with the proposed method are promising. (C) 2000 Elsevi
er Science B.V. All rights reserved.