ETSR-YOLO: An improved multi-scale traffic sign detection algorithm based on YOLOv5.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Autonomous Systems

Background:

• Current traffic sign recognition methods struggle with ambient light, varying target sizes, and complex backgrounds, impacting driverless technology accuracy.
• Existing algorithms often exhibit reduced recognition performance in real-world traffic scenarios.

Purpose of the Study:

• To develop an optimized traffic sign recognition algorithm, ETSR-YOLO, to overcome limitations of current methods in driverless technology.
• To enhance the accuracy and robustness of traffic sign detection under diverse environmental conditions.

Main Methods:

• The study proposes ETSR-YOLO, an optimized algorithm based on YOLOv5s, featuring an enhanced path aggregation network (PANet) for improved multi-scale feature fusion and small object recognition.
• Incorporation of two improved C3 modules to suppress background noise and boost feature extraction capabilities.
• Integration of the Wise-IoU (WIoU) function for enhanced learning ability and sample robustness.

Main Results:

• ETSR-YOLO achieved a 6.6% mAP@0.5 improvement on the TT100K dataset and a 1.9% improvement on the CCTSDB2021 dataset.
• The algorithm demonstrated a short average inference time on an embedded computing platform, suitable for real-time applications.
• Experimental results confirm ETSR-YOLO's dependable performance in real-world traffic scenes.

Conclusions:

• ETSR-YOLO significantly improves traffic sign recognition accuracy and robustness, addressing key challenges in driverless technology.
• The optimized algorithm offers efficient and reliable traffic sign detection for intelligent vehicles.
• The study provides a valuable contribution to the advancement of autonomous driving systems.