Road Crack Detection
This project aims to develop a computer vision system for automatically detecting and classifying various types of road cracks. This system can be used to improve road maintenance efficiency and safety by enabling faster and more objective identification of road damage. This project utilizes the YOLOv8m deep learning model. The "m" variant signifies the smaller and faster version of the YOLOv8 model.
Dataset Classes:
The system is trained to recognize four critical road defect categories:
- Crocodile Cracks: These wide, alligator-like cracks indicate severe pavement distress and require immediate repair.
- Longitudinal Cracks: Cracks running parallel to the direction of traffic flow, often caused by fatigue or shrinkage in the asphalt.
- Potholes: Circular depressions in the road surface that pose a hazard to vehicles and can worsen if left unattended.
- Transverse Cracks: Cracks running perpendicular to the traffic flow, typically caused by traffic loading or environmental factors.
Datasets:
To train the YOLOv8m model effectively, the project leverages two publicly available road crack datasets:
- Roboflow - https://universe.roboflow.com/road-crack-project/road-crack-detection-combined-dataset
- Roboflow - https://universe.roboflow.com/search?q=class%3Ahelp
Results:
Class Images Instances Box(P R mAP50 mAP50-95): 100% 6/6 [00:06<00:00, 1.03s/it]
all 82 248 0.633 0.589 0.61 0.309
Crocodile 82 103 0.77 0.495 0.658 0.341
Longitudinal 82 71 0.529 0.549 0.484 0.171
Pothole 82 25 0.783 0.68 0.752 0.515
Transverse 82 49 0.45 0.633 0.546 0.211