Optical flow estimation using brox

Abstract

This thesis introduces a process for computing optical flow between two successive images with the variational model suggested by Brox et al. There are three assumptions to be satisfied: grey value constancy, gradient constancy, and smoothness. The energy functional consists of these assumptions with weight parameters. Since the energy functional to be minimized and the corresponding Euler-Lagrange equation are highly nonlinear, several linearization processes such as Taylor expansion and fixed point iterations are performed. In addition, the coarse-to-fine strategy called pyramidal scheme approach is applied in order to handle large displacement flow. Since the choice of weight parameters of each assumption and the stop condition of each iteration might cause different results, we test various experiments and compare those results.