A Study on Robust Object Detection Models for Real-World Environments Using Generative AI-Based Augmentation

Recent advances in deep learning–based object detection have significantly improved scene understanding; however, detection performance in road traffic environments is still limited by illumination changes, adverse weather, and background complexity. Challenging conditions—such as rain, fog, and nighttime—introduce severe visual domain shifts that reduce the reliability of autonomous driving and traffic surveillance systems. To address this issue, this study proposes a style-driven data augmentation framework leveraging GAN-based generative AI techniques. The method synthesizes realistic atmospheric and illumination conditions beyond conventional photometric adjustments, enriching dataset variability and improving model generalization. Experimental results show that models trained with the augmented dataset consistently outperform baselines trained only on raw data, confirming the effectiveness of appearance-level generative augmentation for robust perception. The proposed framework demonstrates strong applicability to autonomous vehicles, Intelligent Transportation Systems (ITS), and smart city infrastructure.