A Robust Algorithm for Optical Flow Estimation

Abstract

Most of the existing methods for optical flow estimation are based on a constraint equation which is defined for each image pixel. This class of algorithms is usually called gradient-based. Due to the structure of the constraint equation, the problem is ill-posed, thus some solutions based on regularization have been proposed in the past. On the contrary, if other constraint equations can be found for the pixel under consideration or in its immediate neighborhood, the problem is not ill-posed and a solution can be found by solving determined or over-determined systems of equations, Following this reasoning, several algorithms for evaluating the optical flow have been proposed in the literature. Most of these over-determined systems of equations are solved by using least-squares techniques, In this paper, a new approach is presented in order to eliminate, or strongly reduce, the drawbacks of least-squares and regularization-based techniques. This is based on a modified version of the Combinatorial Hough Transform. A comparison is made between the results obtained with the new approach and those produced by the classical least-squares and regularization-based techniques.