A Hierarchical Approach for Traffic Sign Recognition Based on Shape Detection and Image Classification

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Autonomous vehicles require robust traffic sign detection and recognition for safe operation.
• Current methods often struggle with real-time performance for fast-moving vehicles.
• A two-phase approach is common, but performance can be further optimized.

Purpose of the Study:

• To develop and evaluate a novel two-phase deep learning system for traffic sign detection and classification.
• To compare the proposed method against the state-of-the-art YOLOv5 detector.
• To improve precision, recall, and mean Average Precision (mAP) in traffic sign recognition.

Main Methods:

• Utilized Mask R-CNN for shape-based traffic sign detection in the first phase.
• Employed the Xception model for traffic sign classification in the second phase.
• Collected and utilized a dataset of 11,074 Taiwanese traffic signs captured via mobile and GoPro cameras.

Main Results:

• The proposed Mask R-CNN and Xception approach demonstrated significant improvements in precision, recall, and mAP.
• Experiments were conducted using both class-based and shape-based dataset versions.
• The method showed superior performance compared to the YOLOv5 detector in conducted tests.

Conclusions:

• The proposed two-phase deep learning strategy effectively addresses traffic sign detection and recognition challenges.
• This approach offers a promising solution for enhancing the safety and reliability of autonomous driving systems.
• Further research can explore real-world deployment and integration into existing autonomous driving platforms.