We describe the problem of relative orientation in terms of homogeneous coordinates concluding in a least squares problem in the observed image coordinates. The solution determines a rotational matrix for each image; these rotational matrices bring the images back to the normal position. The explicit formula for the rotational matrices is derived using properties of ‘nearly’ orthogonal matrices. The procedure is augmented by a special preliminary iteration step in order to cope with large rotations.
The method is described through a complete Pascal program.