VISIBILITY ENCHANCEMENT AND OPTICAL FLOW ESTIMATION UNDER ADVERSE WEATHER CONDITIONS

Abstract

The goal of this thesis is to develop advanced visibility enhancement methods and optical flow algorithms for outdoor vision-based systems under a range of dynamic bad weather phenomena, especially the rainy condition. Rainy scenes mainly produce two problematic effects for vision-based algorithms that arising from rain streaks and from rain veiling, which is the term we use to denote the effect. Most of existing optical flow algorithms do not perform adequately due to the presence of these two effects. In this thesis, we will present a series of works to both solve single image deraining and to compute optical flow in the rainy scenes over multiple image frames. Towards the end of the thesis, we will present a universal architecture that can do visibility enhancement for a range of bad weather phenomena.