CSW-YOLO: A traffic sign small target detection algorithm based on YOLOv8.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Autonomous Systems

Background:

• Real-time traffic sign detection is crucial for autonomous driving safety.
• Existing algorithms struggle with small targets in complex traffic scenarios.
• YOLOv8 offers a foundation but requires optimization for specific tasks.

Purpose of the Study:

• To develop a more efficient and accurate small target detection algorithm for traffic signs.
• To enhance the YOLOv8 model for improved real-time performance in autonomous driving.
• To reduce computational load while maintaining high detection precision.

Main Methods:

• Modified the YOLOv8 C2f module with FasterNet's residual Faster-Block and TransNeXt's CGLU (C2f-faster-CGLU).
• Integrated large separable kernel attention (LSKA) with SPPF (SPPF-LSKA) for enhanced feature extraction.
• Incorporated a dedicated small target detection layer and the WISE-Inner-MPDIoU loss function.

Main Results:

• Achieved Map50 scores of 89.8% on TT100K and 98.9% on CCTSDB 2021 datasets.
• Demonstrated precision comparable to mainstream algorithms.
• Significantly reduced model parameters and computational requirements.

Conclusions:

• The proposed CSW-YOLO model effectively improves traffic sign detection for autonomous driving.
• The enhancements lead to better accuracy and efficiency, particularly for small targets.
• The model is well-suited for real-time applications with reduced computational demands.