Quantification method of concrete pavement diseases based on rotating box annotation.

Area of Science:

• Civil Engineering
• Transportation Infrastructure
• Computer Vision

Background:

• Concrete pavement condition impacts road safety and ride comfort.
• Current horizontal box algorithms for pavement disease detection have limitations in accurate dimension quantification and duplicate detection.
• Intelligent quantification of pavement diseases is crucial for effective road maintenance.

Purpose of the Study:

• To address limitations in current pavement disease detection methods.
• To develop an improved system for accurate quantification and enumeration of concrete pavement diseases.
• To establish a robust foundation for automated pavement condition assessment.

Main Methods:

• Established an unmanned aerial vehicle-based concrete pavement disease dataset with rotated-box annotations.
• Validated the impact of dataset split and network architecture on model detection accuracy.
• Conducted comparative experiments for disease dimension quantification (model vs. manual detection).
• Evaluated the impact of target detection models on tracking outcomes using an enhanced model.

Main Results:

• Disease size quantification achieved a relative error of less than 5.42%.
• Disease enumeration precision reached 90%, a 44% improvement over the baseline model.
• The enhanced model demonstrated robust performance in detecting and quantifying pavement diseases.

Conclusions:

• The developed model offers a dependable solution for accurate concrete pavement disease quantification.
• The improved detection and enumeration precision contribute to enhanced road safety and maintenance efficiency.
• This research provides a foundation for advanced automated pavement condition assessment systems.