UAV Dataset for Automated Road Surface Degradation Detection in Real-World Conditions

The RoadAnomaly-YOLO Dataset is a curated collection of 11,024 high-resolution RGB images (640×640 px) designed for road surface anomaly detection using deep learning. The images were captured using UAVs (drones) and ground-based cameras across diverse urban and rural environments, under varying lighting and weather conditions.

Each image is annotated using YOLO-format bounding boxes, enabling immediate integration with YOLOv5/YOLOv8/Ultralytics training pipelines etc. Anomalies may appear once or multiple times per image.

Dataset Statistics

⦁ Total images: 11,024
⦁ Split:
⦁ Train: 8,306 images
⦁ Validation: 2,012 images
⦁ Test: 706 images
⦁ Image resolution: 640 × 640 px (uniform)
⦁ Image type: RGB (JPEG/PNG)
⦁ Annotation format: YOLO TXT (class_id x_center y_center width height)

Classes Included

The dataset covers eight road anomaly categories, widely used in pavement inspection research:

⦁ Alligator Cracking
⦁ Longitudinal Crack
⦁ Transverse Crack
⦁ Rutting
⦁ Pothole
⦁ Stripping
⦁ Raveling
⦁ Bleeding

These anomalies were manually annotated using bounding boxes.

Purpose and Applications

The dataset is intended to support:

⦁ UAV-based road condition monitoring
⦁ Intelligent transportation systems
⦁ Automated maintenance planning
⦁ Object detection model training (YOLO, Faster R-CNN, RetinaNet, etc.)
⦁ Benchmarking and academic research

It enables both training and real-time inference for detecting pavement distresses in real-world environments.