RT Journal Article
T1 One Possible Path Towards a More Robust Task of Traffic Sign Classification in Autonomous Vehicles Using Autoencoders
A1 Martinović, Ivan
A1 Mateo Sanguino, Tomás Jesús
A1 Jovanović, Jovana
A1 Jovanovic, Mihailo
A1 Djukanović, Milena
AB The increasing deployment of autonomous vehicles (AVs) has exposed criticalvulnerabilities in traffic sign classification systems, particularly against adversarial attacksthat can compromise safety. This study proposes a dual-purpose defense frameworkbased on convolutional autoencoders to enhance robustness against two prominent whiteboxattacks: Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD).Experiments on the German Traffic Sign Recognition Benchmark (GTSRB) dataset showthat, although these attacks can significantly degrade system performance, the proposedmodels are capable of partially recovering lost accuracy. Notably, the defense demonstratesstrong capabilities in both detecting and reconstructing manipulated traffic signs, evenunder low-perturbation scenarios. Additionally, a feature-based autoencoder is introduced,which—despite a high false positive rate—achieves perfect detection in critical conditions,a tradeoff considered acceptable in safety-critical contexts. These results highlight thepotential of autoencoder-based architectures as a foundation for resilient AV perceptionwhile underscoring the need for hybrid models integrating visual-language frameworksfor real-time, fail-safe operation.
PB MDPI
SN 2079-9292 (electrónico)
YR 2025
FD 2025-06
LK https://hdl.handle.net/10272/25713
UL https://hdl.handle.net/10272/25713
LA eng
NO Martinović, I., Mateo Sanguino, T. d. J., Jovanović, J., Jovanović, M., & Djukanović, M. (2025). One Possible Path Towards a More Robust Task of Traffic Sign Classification in Autonomous Vehicles Using Autoencoders. Electronics, 14(12), 2382. https://doi.org/10.3390/electronics14122382
DS Repositorio Institucional de la Universidad de Huelva
RD 1 jun 2026