Optimization methods for l 1 -energy minimization in the estimation of optical flow

Abstract

The aim of this paper is to present a new approach for solving the minimization problem for a large class of energy functionals that appear in the differential models of optical flow estimation problems, and which are expressed using the discrete l1-norm. The choice of l1-energy minimization is motivated by the fact that quadratic l2 optimization is not robust to outliers and that l1-norm is a better choice for modeling real problems involving discrete signals. The method described in this paper is very general, thus the advantage of being applicable to almost every differential model that has been proposed so far for the optical flow estimation problem. In order to test and validate our method, a MATLAB implementation on several optical flow models is currently under development. Also, a multi-core implementation on GP-GPU is to be considered in the near future.