ClearSight: a unified model to enhance traffic sign visibility under adverse weather conditions

Abstract

Advanced Driver Assistance Systems (ADAS) often struggle in adverse weather conditions such as rain, fog, and low light, impairing traffic and road sign detection and recognition. Existing research often addresses these conditions separately, leading to models that falter when multiple weather phenomena co-occur. Unlike unified models that primarily focus on fog and rain and may introduce artifacts, this paper presents a novel algorithm, ClearSight. ClearSight removes weather degradation from images, enhancing road sign detection under adverse weather conditions. Initially, machine-learning algorithms were used to extract features from various weather conditions and trained to mitigate weather degradation. ClearSight then integrates these trained algorithms into a unified platform, effectively addressing rain, fog, and low light within a single framework.The proposed framework processes degraded images, producing clear images for accurate road sign detection. ClearSight underwent comprehensive evaluation against state-of-the-art enhancement models, achieving the highest average traffic sign recognition confidence score of 0.91. Addition ally, ClearSight’s enhanced images showed a significant reduction in BRISQUE scores compared to input weather-degraded images. These results demonstrate ClearSight’s superiority over current models, offering enhanced image quality and reliable traffic sign detection in adverse weather conditions. This unified approach for multiple weather conditions significantly enhances ADAS effectiveness in poor visibility scenarios.