Exploring Deep Learning Architectures for Enhanced Traffic Sign Recognition: A Comparative Analysis
Amit Kumar Gupta, Priya Mathur, Riya Singh, Sana Sharma · 2024
This research paper investigates the efficacy of various deep learning architectures in Traffic Sign Recognition (TSR) tasks, aiming to contribute to the advancement of intelligent transportation systems. A comprehensive comparative analysis is conducted, evaluating the performance of four prominent architectures: AlexNet, DenseNet, VGG, and a generic CNN, using the GTSRB dataset. Results demonstrate that DenseNet achieves the highest accuracy of 95.12%, outperforming other architectures, with remarkable precision, recall, and F1-score values. The generic CNN also exhibits impressive performance, with an accuracy of 97.31% and comparable metrics. Conversely, AlexNet and VGG achieve accuracies of 88.84% and 85.03%, respectively, with varying performance metrics. These findings underscore the effectiveness of DenseNet and the generic CNN in accurately classifying traffic signs, highlighting their potential for real-world implementation. The study emphasizes the importance of selecting appropriate deep learning architectures tailored to specific tasks, and further analysis may explore strategies to optimize model performance and generalization capability.