A VISIBILITY LEVEL ESTIMATION VIA LATE-FUSION BASED ON PROBABILITIES OF MULTIPLE CLASSIFIERS USING IN-VEHICLE CAMERA VIDEOS UNDER NIGHTTIME

Abstract

It is necessary to provide visibility information to road users since poor visibility due to snowstorms happens and increase the risk of transportation accidents in winter roads. In particular, visibility conditions are likely to change by the influence of snowstorms, headlights, and road lighting in winter road under nighttime. In this paper, a novel visibility level estimation utilizing invehicle camera videos under nighttime is proposed. In the proposed method, five types offeatureswhich areuseful for estimating visibility level under nighttimeare obtained. In addition, multipleclassifiers are constructed based on the features. Finally, visibility level is estimated by fusing the probability of each classifier. By utilizing the proposed method, we expect to reduce the risk of transportation accidents since the visibility condition can be estimated and provided automatically. In the last of this paper, we verify the effectiveness of the proposed method by the experiments used the actual videos under nighttime.