YOLO-ERCD: An Upgraded YOLO Framework for Efficient Road Crack Detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Transportation Engineering

Background:

• Road damage detection is vital for intelligent transportation systems.
• Current models struggle with fine cracks, lighting variations, and complex backgrounds.
• Existing visual sensing technologies require enhanced accuracy and robustness.

Purpose of the Study:

• To propose YOLO-ERCD, an enhanced YOLO-based framework for improved road crack detection.
• To address limitations in feature representation, lighting adaptation, and background interference in existing models.
• To enhance accuracy and robustness for automated road inspection using sensor-acquired images.

Main Methods:

• Implemented a residual convolutional block attention module for feature representation.
• Integrated a channel-wise adaptive gamma correction module for lighting robustness.
• Developed a visual focus noise modulation module to reduce background interference.
• Utilized datasets from vehicle-mounted and traffic surveillance cameras.

Main Results:

• YOLO-ERCD demonstrated superior accuracy and computational efficiency compared to recent models.
• The framework effectively addresses challenges in fine crack detection and lighting variations.
• Experimental results validated performance on both proprietary and public datasets.

Conclusions:

• YOLO-ERCD offers a robust and efficient solution for road damage detection.
• The lightweight design enables real-time deployment on edge devices.
• The study highlights the potential of AI-based visual sensing for advanced road monitoring.