Evaluating the Impact of Adversarial Attacks on Traffic Sign Classification using the LISA Dataset

2025-09-08

AI Summary

Adversarial robustness of traffic sign classifiers remains underexplored on the LISA dataset, posing critical safety risks for autonomous driving systems. Method: This study systematically evaluates the vulnerability of 47-class traffic sign classification models—built upon convolutional neural networks (CNNs)—to adversarial attacks on the LISA dataset. We employ two canonical white-box attacks, Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD), to generate adversarial examples and quantitatively analyze the impact of perturbation magnitude on classification accuracy. Results: Even imperceptible perturbations cause substantial accuracy degradation, exposing severe security vulnerabilities in real-world traffic scenarios. This work establishes the first reproducible benchmark for adversarial robustness assessment on the LISA dataset, bridging a key gap in the literature. Moreover, it provides empirically grounded insights for safety-critical risk modeling and informs the design of robust defense mechanisms for vision-based autonomous driving systems.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Adversarial attacks pose significant threats to machine learning models by introducing carefully crafted perturbations that cause misclassification. While prior work has primarily focused on MNIST and similar datasets, this paper investigates the vulnerability of traffic sign classifiers using the LISA Traffic Sign dataset. We train a convolutional neural network to classify 47 different traffic signs and evaluate its robustness against Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD) attacks. Our results show a sharp decline in classification accuracy as the perturbation magnitude increases, highlighting the models susceptibility to adversarial examples. This study lays the groundwork for future exploration into defense mechanisms tailored for real-world traffic sign recognition systems.

Problem

• Investigating traffic sign classifier vulnerability to adversarial attacks
• Evaluating robustness against FGSM and PGD attack methods
• Assessing accuracy decline with increasing perturbation magnitude

Innovation

• Used LISA dataset for traffic sign classification
• Applied FGSM and PGD adversarial attack methods
• Evaluated CNN robustness against adversarial perturbations