Lane Detection in Foggy Images using Generative Adversarial Networks

Abstract

autonomous vehicles (AVs) and advanced driver assistance systems (ADAS). These adverse conditions compromise the reliability of critical perception tasks such as lane detection and road environment understanding, thereby increasing the risk to driving safety and stability. In this study, a Generative Adversarial Network (GAN)-based approach was developed to address these issues using one of the leading techniques currently employed in the literature. Using GAN models, realistic foggy road images with varying density levels were generated, and lane detection performance was evaluated on this synthetic dataset. A comparative analysis was conducted between classical image processing techniques and deep learning-based methods. The effectiveness of each approach was evaluated using the Intersection over Union (IoU) metric, which balances both accuracy and spatial coverage in region-based tasks such as lane detection. While classical methods achieved an IoU performance of 89%, deep learning-based techniques reached up to 96%. The results demonstrate that deep learning approaches significantly outperform classical methods in identifying complex road structures, particularly under foggy conditions. These findings highlight the potential of GAN-based data generation and deep learning models to enhance the robustness of perception systems under adverse weather conditions, thereby contributing to safer and more reliable autonomous driving.