High Accuracy Homography Computation without Iterations

We present a highly accurate least-squares (LS) alternative to the theoretically optimal maximum likelihood (ML) estimator for homographies between two images. Unlike ML, our estimator is non-iterative and yields a solution even in the presence of large noise. By rigorous error analysis, we derive a “hyperLS” estimator which is unbiased up to second order noise terms. We also introduce a computational simplification, which we call “Taubin approximation”, without incurring an accuracy loss. We experimentally demonstrate that our estimator far surpasses the standard LS and is nearly comparable to the ML and the theoretical accuracy limit (the KCR lower bound).