Two views, eight points, a linear system
On 10 September 1981 Christopher Longuet-Higgins published in Nature an algorithm that recovers the three-dimensional structure of a scene and the relative placement of two viewpoints from point correspondences alone. Eight pairs of corresponding points give a system of linear equations, and that is enough — with nothing known about where the cameras stood.
Why it matters
The relative placement of two cameras became a quantity to be computed rather than measured. Until then two-view geometry meant a nonlinear search needing an initial guess; here it is solved directly, and all later multiple-view geometry rests on it, from structure from motion to visual SLAM.
The paper describes itself this way: when the spatial relationship between the two projections is unknown, reconstructing the scene reduces to finding the relative orientation of the two viewpoints, given that the correspondence problem has been solved. The author is at the Laboratory of Experimental Psychology, University of Sussex. Confidence is medium for one reason: the full text is paywalled and the downloadable PDF is an image scan, as is every copy that could be found. The number eight was not read off the Nature page but from Richard Hartley's peer-reviewed paper "In Defence of the 8-point Algorithm", which states that the algorithm was introduced by Longuet-Higgins and that with 8 point matches known, the solution of a set of linear equations is involved. That paper's reference list independently gives Nature 293:133-135, Sept 1981. The pages are worth a separate note. The common citation of this work as Nature 293, pages 61-62, does not point at this article: it is on pages 133-135, and that is the address at which the Nature page opens.