This project presents a hybrid pothole detection system that combines Digital Image Processing (DIP) techniques with YOLOv8 deep learning detection. It supports pothole identification from images, videos, and live camera feed, while also allowing visualization of each processing stage through a GUI.
The system demonstrates both traditional image processing pipelines and modern object detection in one unified application.
-
Upload image or video for analysis
-
Step-wise visualization of DIP pipeline stages
-
Pothole detection on:
- Images
- Videos
- Live camera feed
-
Custom-trained YOLOv8 model support
-
Modular and scalable project architecture
-
GUI interface built with Tkinter
The YOLOv8 model was trained on Google Colab using a custom pothole dataset.
Training workflow:
- Dataset annotation and conversion to YOLO format
- Training with Ultralytics YOLOv8
- Exporting best weights (
best.pt) - Downloading trained model for local inference
After training, place the model file here:
pothole-detection/models/best.pt
pothole-detection/
│
├── main.py
├── gui/
├── processing/
├── detection/
├── utils/
├── models/
│ └── best.pt
├── screenshots/
├── requirements.txt
└── README.md
git clone https://github.com/avxway/pothole-detection-system.git
cd pothole-detection-system
pip install -r requirements.txt
python main.py
The GUI window will open.
- Click Upload Image/Video
- Explore each DIP processing stage
- Run detection on image, video, or live camera
- Press q to stop video/camera detection
- Python
- OpenCV
- NumPy
- Matplotlib
- Tkinter
- Ultralytics YOLOv8
- PyTorch
- Web dashboard deployment
- Mobile detection system
- Road damage severity estimation
- Automated pothole reporting to authorities
Anmol Verma
This project is created for academic and research purposes.