Single Image-Based Scene Visibility Estimation

Abstract

This paper proposes a novel method to utilize a single daylight image of the observed scene to estimate the extinction coefficient of the observed atmosphere and obtain the scene visibility. The proposed method consists of three steps. First, based on the theory of atmospheric physics, we estimate the extinction coefficient of the clear atmosphere in an observed scene. Second, a method combining the dark channel prior and the edge collapse-based transmission refinement is employed to calculate the ratio between the extinction coefficient of the observed atmosphere and that of the clear atmosphere. This ratio indicates the increased degree of the extinction coefficient due to the increment of the atmospheric turbidity. Finally, by multiplying the extinction coefficient of the clear atmosphere and the ratio, the extinction coefficient of the observed atmosphere is calculated and the scene visibility is obtained. The numerous experiments on the proposed method suggest that it performs well in measuring the scene visibility in various types of scenes without additional assistance (e.g., geometric calibration of the camera, road marking extraction, and ground truth data collection).