Real-time fog visibility range estimation for autonomous driving applications

In this paper, we present a novel fog visibility range estimation algorithm for autonomous driving application. The proposed method is based on a hybrid neural network for which localized image entropy and image-based features are given as an input. While entropy used for visibility estimation, image-based features act as a basis for distance calibration. The proposed network is tested on real data collected and calibrated on-road and performs very well with respect to accuracy. The proposed algorithm is also of low complexity and provides the result on a near real-time basis.