UAV-PDD2023: A benchmark dataset for pavement distress detection based on UAV images

Area of Science:

• Civil Engineering
• Computer Vision
• Remote Sensing

Background:

• Automated pavement distress detection is crucial for infrastructure maintenance.
• Existing datasets may lack diversity in weather conditions, road types, and distress categories.
• Unmanned Aerial Vehicles (UAVs) offer a cost-effective and efficient method for pavement data acquisition.

Purpose of the Study:

• To introduce the UAV-PDD2023, a comprehensive dataset for pavement distress detection.
• To provide a benchmark for evaluating deep learning models in pavement inspection.
• To facilitate research in automated road condition assessment using UAV imagery.

Main Methods:

• Collected over 11,150 pavement distress images using UAVs in China.
• Captured data under diverse weather conditions and across various road types (highways, provincial, county roads).
• Annotated six common pavement distress types: longitudinal cracks, transverse cracks, oblique cracks, alligator cracks, patching, and potholes.

Main Results:

• The UAV-PDD2023 dataset encompasses a wide range of pavement distresses and environmental variations.
• The dataset is suitable for training and validating deep learning models for object detection and image classification tasks.
• It serves as a robust benchmark for assessing the performance of pavement distress detection algorithms.

Conclusions:

• The UAV-PDD2023 dataset significantly advances the field of automated pavement management.
• It enables the development of more accurate and reliable systems for road surface monitoring.
• The dataset is freely available to the research community to foster innovation in pavement engineering.