EULPR: European License Plate Recognition
EULPR is a computer-vision model architecture purpose-built for detecting, reading, and recognizing European license plates. It is optimized for speed and accuracy across diverse EU plate formats.
Model Performance
- Detection Rate: 100.0%
- Text Extraction Rate: 100.0%
- Processing Speed: 7.6 FPS
- Model Size: YOLOv12 Nano (~10.5MB)
Supported Languages
- English (en)
- German (de)
- French (fr)
- Spanish (es)
- Italian (it)
- Dutch (nl)
Quick Start
Installation
pip install ultralytics easyocr opencv-python pillow torch torchvision huggingface_hub
Usage
import cv2
import numpy as np
from ultralytics import YOLO
import easyocr
from PIL import Image
from huggingface_hub import hf_hub_download
import warnings
# Suppress warnings
warnings.filterwarnings('ignore')
# Download models from HuggingFace
print("Downloading model from HuggingFace...")
model_path = hf_hub_download(repo_id="0xnu/european-license-plate-recognition", filename="model.onnx")
config_path = hf_hub_download(repo_id="0xnu/european-license-plate-recognition", filename="config.json")
# Load models with explicit task specification
yolo_model = YOLO(model_path, task='detect')
ocr_reader = easyocr.Reader(['en', 'de', 'fr', 'es', 'it', 'nl'], gpu=False, verbose=False)
# Process image
def recognize_license_plate(image_path):
# Load image
image = cv2.imread(image_path)
image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
# Detect license plates
results = yolo_model(image_rgb, conf=0.5, verbose=False)
plates = []
for result in results:
boxes = result.boxes
if boxes is not None:
for box in boxes:
# Get coordinates
x1, y1, x2, y2 = box.xyxy[0].cpu().numpy()
# Crop plate
plate_crop = image_rgb[int(y1):int(y2), int(x1):int(x2)]
# Extract text
ocr_results = ocr_reader.readtext(plate_crop)
if ocr_results:
text = ocr_results[0][1]
confidence = float(ocr_results[0][2]) # Convert to native Python float
plates.append({'text': text, 'confidence': confidence})
return plates
# Usage Example
results = recognize_license_plate('sample_car_with_license.jpeg')
print(results)
Model Architecture
Detection Model (YOLOv12n)
- Architecture: YOLOv12 Nano
- Parameters: ~3M
- Input Size: 640x640 pixels
- Output: Bounding boxes for license plates
OCR Model (EasyOCR)
- Engine: Deep learning-based OCR
- Languages: Multi-European language support
- Character Set: Alphanumeric + common symbols
Training Details
- Dataset: European License Plate Dataset (0xnu/european-licence-plate)
- Training Epochs: 30
- Batch Size: 16
- Image Size: 640x640
- Optimizer: AdamW
- Framework: Ultralytics YOLOv12
Use Cases
- Traffic monitoring systems
- Automated parking management
- Law enforcement applications
- Toll collection systems
- Vehicle access control
Limitations
- Optimized for European license plate formats
- Performance may vary with extreme weather conditions
- Requires good image quality for optimal text recognition
- Real-time performance depends on hardware capabilities
License
This project is licensed under the Modified MIT License.
Citation
If you use this model in your research or product, please cite:
@misc{eulpr2025,
title={EULPR: European License Plate Recognition},
author={Finbarrs Oketunji},
year={2025},
publisher={Hugging Face},
howpublished={\url{https://huggingface.co/0xnu/european-license-plate-recognition}}
}
Copyright
Copyright (C) 2025 Finbarrs Oketunji. All Rights Reserved.