Nighttime Visibility Classification Based on Stable Light Sources

Abstract

To enhance the accuracy of existing nighttime visibility estimation methods, this study proposes a classification algorithm for nighttime visibility levels based on stable light sources. Initially, a target detection network identifies all stable streetlights in the image and extracts the light source blocks. Subsequently, these blocks undergo fog classification through a classification network. The blocks we then sorted by brightness values and assigned corresponding weights. Finally, the classification results and weights are combined to categorize the nighttime image visibility levels. Experimental results show that the accuracy of our nighttime visibility classification algorithm reaches 77.6% on real-world datasets, outperforming existing methods and demonstrating good generalization across different scenes.