Road potholes are one of the major causes of road accidents and vehicle damage. Manual inspection of roads is time-consuming and inefficient. This project proposes an automated pothole detection system using Image Processing and Deep Learning techniques.
The system analyzes road images captured through cameras or drones and identifies potholes using YOLOv8 object detection models.
- Detect potholes automatically from road images.
- Reduce manual road inspection efforts.
- Improve road safety through early pothole detection.
- Support smart city infrastructure management.
- Python
- OpenCV
- YOLOv8
- PyTorch
- TensorFlow
- Computer Vision
- Image Processing
- Machine Learning
Road images containing potholes were collected from publicly available datasets and road damage image repositories.
- Image resizing
- Image normalization
- Annotation conversion to YOLO format
- YOLOv8 pre-trained model
- Transfer learning
- Object detection training
Performance was measured using:
- Precision
- Recall
- F1 Score
- Mean Average Precision (mAP)
Road Image ↓ Image Preprocessing ↓ Feature Extraction ↓ YOLOv8 Model ↓ Pothole Detection ↓ Bounding Box Output
The system successfully detected potholes from road surface images with high accuracy.
Benefits:
- Faster road inspection
- Reduced maintenance cost
- Improved road safety
- Scalable infrastructure monitoring
- Real-time pothole detection
- Drone-based road monitoring
- Smart city integration
- GPS-enabled pothole reporting