Dark Channel Based Visibility Measuring from Daytime Scene Videos

Abstract

Visibility is one of the most important factors of meteorological observation. Low visibility caused by fog often has a great impact on human production and daily life. Therefore, visibility measurements and warnings are given as early as possible to avoid accidents. This paper proposes a new visibility measuring method based on dark channel characteristics of digital images in natural scenes. It uses the means of dark channel values obtained from foggy images and high visibility reference images as inputs of a deep fully connected neural network, and then the trained model is applied to estimate visibility. Experiments conducted on a dataset contained two foggy scenes. The results demonstrate the effectiveness of our proposed method for daytime visibility measurement. At the same time, this paper proposes and designs an automatic detection system based on our algorithm. It implements the real time visibility measuring through edge computing.