VISIBILITY ENHANCEMENT AND OPTICAL FLOW ESTIMATION IN EXTREME OUTDOOR CONDITIONS

Abstract

In this thesis, our primary goal is to develop novel visibility enhancement and optical flow estimation methods from images and videos which are degraded by rain. The degradation caused by rain mainly has three aspects: rain streaks, the veiling effect caused by rain accumulation, and raindrops. In this thesis, we firstly separate different kinds of degradation caused by rain into different chapters and then provide solutions respectively. Next, we focus on two kinds of extension problems in the following chapters: the extension of robust optical flow estimation for rainy data, and the extension of our deraining method from daytime condition to nighttime. In both of these two extensions, raindrop and rain streak can be simplified and we only need to consider the rain accumulation. Therefore, the two extension problems can also be equivalent to optical flow estimation for foggy data and nighttime fog removal.