Visibility estimation via near-infrared bispectral real-time imaging in bad weather

This manuscript proposes a physical inversion based visibility estimation model. The proposed approach uses a dual-spectral imaging system, which is composed of near-infrared cameras, lenses, filters, condenser, and spectroscope. The visibility is estimated using the two images captured over the selected near-infrared wavelengths. In order to increase the precision of estimated visibility, a visibility error correction model based on a machine learning regression algorithm is proposed. We address the issue of accurate visibility estimation, which is essential for various applications in environmental conservation, meteorological monitoring, and industrial pollutant emission surveillance. The results of experiments, which have conducted in this study, validated the effectiveness of our proposed method in visibility estimation in comparison with the existing methods.