Atmospheric visibility estimation: a review of deep learning approach

Estimating atmospheric visibility is a critical task in our current time, particularly in the safety of aviation, navigation, and highway traffic, because it measures the quality of the atmosphere’s transparency. Poor visibility caused by climatic factors such as fog, haze, and clouds makes driving difficult and can lead to serious road accidents. Therefore, the researchers proposed several Deep Learning (DL) approaches to facilitate visibility estimation instead of using meteorological equipment, which is very expensive. This paper aims to give an overview of the DL systems used in previous studies to estimate atmospheric visibility. These systems are categorized based on the input data type, with some based on tabular data and others on image data. The methodologies, datasets, and Deep Neural Network (DNN) architectures used in the reviewed studies are thoroughly discussed. The evaluation metrics are also compared. Finally, the common limitations of the proposed approaches are highlighted, as well as potential future trends in this research field.