Application of convolutional neural networks for visibility estimation of CCTV images

In this paper we discuss the possibility of application of a Convolutional Neural Network for visual atmospheric visibility estimation. A system utilizing such a neural network can greatly benefit a smart city by providing real time localized visibility data across all highways and roads by utilizing a dense network of traffic and security cameras that exist in most developed urban areas. To achieve this, we implemented a Convolutional Neural Network with 3 convolution layers and trained it on a data set taken from CCTV cameras in South Korea. This approach allowed us achieve accuracy above 84%. In the paper we describe the network structure and training process, as well as some final thoughts on the next steps in our research.