Motion estimation using second order gradient methods

Gradient based motion estimation techniques (GM) are considered to be in the heart of state-of-the-art registration algorithms, being able to account for both pixel and subpixel registration and to handle various motion models (translation, rotation, affine, projective). These methods estimate the motion between two images using a first order Taylor expansion of the registered images. This paper offers two main contributions: First, we improve the convergence rate and range of the GM using a formulation based on a second order Taylor expansion. Second, we extend the second order algorithm using a symmetrical formulation which further improves the convergence properties. The results are verified by theoretical analysis and experimental trials.