ESA-YOLO: An efficient scale-aware traffic sign detection algorithm based on YOLOv11 under adverse weather conditions

Area of Science:

• Computer Vision
• Artificial Intelligence
• Autonomous Driving Systems

Background:

• Traffic sign detection is crucial for autonomous driving and driver assistance systems.
• Existing methods struggle with multi-scale, small objects, and complex/adverse conditions.
• High accuracy and efficiency remain key challenges.

Purpose of the Study:

• To propose an improved YOLOv11-based algorithm for enhanced traffic sign detection.
• To address challenges in detecting multi-scale, small objects, and improving robustness in complex scenes.
• To achieve superior performance with comparable computational complexity.

Main Methods:

• Developed a Dense Multi-path Feature Pyramid Network (DMFPN) for effective multi-scale feature fusion.
• Introduced a Context-Aware Gating Block (CAGB) to integrate local and global context for small object detection.
• Implemented an Adaptive Scene Perception Head (ASPH) combining multi-scale features and attention for robustness.

Main Results:

• Outperformed state-of-the-art YOLOv11n on TT100K dataset (3.8% mAP@50, 3.9% mAP@50-95).
• Achieved significant gains on CCTSDB2021 dataset (2.3% mAP@50, 1.8% mAP@50-95).
• Demonstrated superior small object detection and robustness in complex environments with reduced parameters (20%).

Conclusions:

• The proposed YOLOv11 improvements offer a more accurate and efficient solution for traffic sign detection.
• The DMFPN, CAGB, and ASPH components effectively enhance multi-scale, small object detection, and robustness.
• The model provides a promising advancement for autonomous driving and driver assistance systems.