GCPOF: a novel optical flow estimation algorithm leveraged by sparse ground control points

Abstract

As a fundamental task in computer vision, optical flow estimation algorithms aim to establish dense pixel correspondences between image frames. This paper presents a novel optical flow estimation framework called GCPOF to handle large displacement and scale variations of scene objects, which appear frequently and pose great challenges in practice. Within the framework of GCPOF, large displacement and scale variations are captured by a new problem formulation leveraged by sparse ground control points. We present detailed theoretical derivation of the solution to the problem based on iterative reweighted least squares. Both qualitative and quantitative evaluations on synthetic and real images demonstrate that GCPOF is able to handle optical flow fields with large displacement and scale variations properly, and it runs significantly faster than relevant optical flow estimation methods.