SeoulTech researchers develop AI framework to test traffic sign recognition

Researchers from Seoul National University of Science and Technology, Kyung Hee University, and the Technical University of Munich have developed a new artificial intelligence framework called Adversarial Wear and Tear (AdvWT). The framework is designed to assess how natural deterioration of traffic signs—caused by sunlight, rain, dirt, and corrosion—can lead to errors in deep neural network-based vision systems.

Unlike temporary optical attacks, natural wear and tear is a persistent issue that can remain until a sign is physically repaired or replaced. The AdvWT framework utilizes a GAN-based image-to-image translation model to learn and reproduce realistic forms of damage. By testing these generated images against various recognition architectures, the researchers aim to identify vulnerabilities in autonomous driving systems and improve AI generalization to real-world environmental conditions.