A Two-Phase Approach for Efficient Traffic Sign Detection and Recognition

Objectives: The objective of this study is to enhance the accuracy of traffic sign detection and recognition in modern intelligent transport systems, addressing real-time challenges under varying conditions. Methods: A two-phase approach is adopted. The first phase employs the You Only Look Once version 8 (YOLOv8) architecture to efficiently detect traffic signs under real-time conditions, considering variables like adverse weather and obstructions. Subsequently, the second phase employs a sequential convolutional network for precise recognition, utilizing the output from the first phase. This integrated method enhances traffic sign detection and recognition, contributing to road safety and efficient traffic management in complex transportation scenarios. Findings: The YOLOv8 architecture, utilized in Phase 1, demonstrated exceptional performance with a mean Average Precision (mAP) of 0.986 during validation. In Phase 2, the Convolutional Neural Network (CNN)-based recognition model achieved an impressive test accuracy of 98.7% on 463 test images, with a low-test loss of 0.1186, indicating consistent accuracy. The robustness of both models is confirmed by successful testing with three unseen images. YOLOv8 accurately detected and classified these images, while the CNN model correctly recognized them. These findings underscore the effectiveness of the two-phase approach in enhancing traffic sign detection and recognition, with significant implications for improving road safety and traffic management in real-world scenarios. Novelty: The novelty of this approach lies in its seamless integration of YOLOv8 for efficient traffic sign detection and a sequential convolutional network for accurate recognition, offering a significant advancement in addressing real-time challenges and contributing to enhancing road safety and traffic management in an increasingly complex transportation landscape. Keywords: Traffic sign detection, Traffic sign recognition, Convolutional Neural Networks, YOLOv8, Object detection