The two view structure is equivalent to the fundamental matrix. Since the fundamental matrix is a matrix determined up to an arbitrary scale factor, 8 equations are required to obtain a unique solution. The simplest way to compute the fundamental matrix consists of using Equation (3.26). This equation can be rewritten under the following form:

(D6) |

This system of equation is easily solved by Singular Value Decomposition (SVD) [43]. Applying SVD to yields the decomposition with and orthonormal matrices and a diagonal matrix containing the singular values. These singular values are positive and in decreasing order. Therefore in our case is guaranteed to be identically zero (8 equations for 9 unknowns) and thus the last column of is the correct solution (at least as long as the eight equations are linearly independent, which is equivalent to all other singular values being non-zero).

It is trivial to reconstruct the fundamental matrix from the solution vector . However, in the presence of noise, this matrix will not satisfy the rank-2 constraint. This means that there will not be real epipoles through which all epipolar lines pass, but that these will be ``smeared out'' to a small region. A solution to this problem is to obtain as the closest rank-2 approximation of the solution coming out of the linear equations.