Nonlinear observer design on SL(3) for homography estimation by exploiting point and line correspondences with application to image stabilization

Abstract

Although homography estimation from correspondences of mixed-type features, namely points and lines, has been relatively well studied with algebraic approaches by the computer vision community, this problem has never been addressed with nonlinear observer paradigms. In this paper, a novel nonlinear observer on the Special Linear group applied to homography estimation is developed. The key advance with respect to similar works on the topic is the formulation of observer innovation that exploits directly point and line correspondences as input without requiring prior algebraic reconstruction of individual homographies. Rigorous observability and stability analysis are provided. A potential application to image stabilization in presence of very fast camera motion, severe occlusion, specular reflection, image blur, and light saturation is demonstrated with very encouraging results.