Seeing Through Bad Weather

Abstract

Current vision systems are designed to perform in clear weather. Needless to say, in any outdoor application, there is no escape from bad weather. Ultimately, computer vision systems must include mechanisms that enable them to function (even if somewhat less reliably) in the presence of haze, fog, rain, hail and snow. We begin by studying the visual manifestations of different weather conditions. For this, we draw on what is already known about atmospheric optics, and identify effects caused by bad weather that can be turned to our advantage; we are not only interested in what bad weather does TO vision but also what it can do FOR vision.

Since the atmosphere modulates the light traveling from a scene point to the observer, it can be viewed as a mechanism of visual information coding. We derive two physics-based models that describe how scene contrasts and colors appear in bad weather. Based on these models, we develop several algorithms for recovering pertinent scene properties, such as three-dimensional structure and clear day contrasts as well as colors, from one or more images taken under poor weather conditions.

To validate our techniques, we acquired a large dataset of high quality images of an outdoor scene every hour for a period of 9 months. This dataset includes images under a wide range of weather and illumination conditions as well as different seasons. Therefore, in addition to vision in bad weather, this database also has implications for computer graphics, remote sensing and atmospheric sciences. The effectiveness of our methods is demonstrated using several experiments with images under harsh weather conditions. We conclude by presenting a discussion on our current and future research in this area.