To solve for h, why wouldn't we just use RREF to solve for the nullspace of A, and that would give a vector that can be scaled up to h?
mpotoole
Using Reduced Row-Echelon Form to find the nullspace would work in this example.
However, in practice, you may be dealing with many correspondences. And because of noise associated with these samples, there may be no $h$ that solves this equation exactly. So it's often useful to use the SVD to find the smallest singular vector instead.
To solve for h, why wouldn't we just use RREF to solve for the nullspace of A, and that would give a vector that can be scaled up to h?
Using Reduced Row-Echelon Form to find the nullspace would work in this example.
However, in practice, you may be dealing with many correspondences. And because of noise associated with these samples, there may be no $h$ that solves this equation exactly. So it's often useful to use the SVD to find the smallest singular vector instead.