Generalized least squares-based parametric motion estimation

The estimation of parametric global motion had a significant attention during the last two decades, but despite the great efforts invested, there are still open issues. The most important ones are related to the accuracy of the estimation and to the ability to recover large deformation between images. s paper, a new generalized least squares-based motion estimator is proposed. The non-linear Brightness Constancy Assumption is directly used instead of using the classical approach by linearizing the minimization problem using the optical flow equation. In addition, the proposed formulation of the motion estimation problem provides an additional constraint that helps to match the pixels by using the image gradient in the matching process. That is achieved by means of a weight for each observation, assigning high weight values to the observations considered as inliers, i.e. the ones that support the motion model, and low values to the ones considered as outliers. The accuracy of our approach has been tested using challenging real images using both affine and projective motion models. Two motion estimator techniques that uses iteratively reweighted least squares-based (IRLS) techniques to deal with outliers, have been selected for comparison purposes. The results obtained show that the proposed motion estimator can obtain, in most cases, more accurate estimates that the IRLS-based techniques.