Malek-Messaoudi/Tunisian-licence-plate-detection

🇹🇳 Tunisian License Plate Detection (YOLOv8n)

This model detects and localizes Tunisian vehicle license plates using the lightweight YOLOv8n architecture.

---

Model Description

• Model Type: YOLOv8n (Nano)
• Task: Object Detection
• Classes: 1 class — "plate"
• Purpose: Detecting Tunisian license plates for ALPR/OCR systems
• Stage: Used as the first step in a complete license plate recognition pipeline

---

Use Case

This model is designed for real-time license plate detection:

• Detect the plate region from vehicle images or video streams
• Crop or send the detected plate to an OCR model
• Improve accuracy for:
  • Parking management
  • Smart surveillance
  • Toll systems
  • Traffic monitoring
  • IoT / edge camera solutions

The lightweight YOLOv8n architecture ensures fast inference even on low-power devices such as Jetson Nano, Raspberry Pi, and mobile CPUs.

---

Training Details

• Base Model: YOLOv8n pretrained on COCO
• Framework: Ultralytics YOLOv8
• Dataset: Custom Tunisian license plate dataset

---

Usage

Python Inference Example

from ultralytics import YOLO

# Load model from Hugging Face Hub
model = YOLO("Safe-Drive-TN/Tunisian-Licence-plate-Detection")

# Run prediction
results = model.predict("path/to/car.jpg", conf=0.5)

# Process results
for result in results:
    boxes = result.boxes
    for box in boxes:
        print(f"Confidence: {box.conf[0]:.2f}")
        print(f"BBox: {box.xyxy[0]}")

Model Files

licence-plate-detection.pt - Best weights from training

last-weights-licence-plate-detection.pt - Last checkpoint

Example

Citation

If you use this model, please cite:

@misc{tunisian-license-plate-detection-2025,
  author       = {Yassine Mhirsi and Malek Messaoudi},
  title        = {Tunisian License Plate Detection},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Malek-Messaoudi/Tunisian-licence-plate-detection}}
}

license

license: mit