Automatic Jordanian License Plate Detection and Recognition System Using Deep Learning Techniques.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Increasing urbanization leads to a surge in vehicle numbers, making manual identification of license plates and logos challenging.
• Automated vehicle identification systems are crucial for efficient traffic management and law enforcement in urban environments.

Purpose of the Study:

• To develop and evaluate a deep learning-based system for the automatic identification of Jordanian vehicles.
• To accurately detect license plates, recognize characters on plates, and identify vehicle logos using transfer learning techniques.

Main Methods:

• Utilized transfer learning with deep learning models: YOLOv3 for license plate and logo detection/character recognition, and VGG16 for logo recognition.
• Trained models on four custom datasets comprising Jordanian vehicle images, license plates, and logos.
• Employed performance metrics including precision, recall, F-measure, and mean average precision (mAP).

Main Results:

• Achieved high accuracy in license plate detection (mAP 99.9%) and character recognition (F-measure 99.95%).
• Demonstrated strong performance in vehicle logo detection (mAP 99.1%) and recognition (F-measure 98%).
• The integrated license plate recognition system achieved 99.8% precision, recall, and F-measure.

Conclusions:

• The developed deep learning system effectively automates the identification of Jordanian vehicles, including license plates and logos.
• The system's high accuracy demonstrates its potential for practical applications in traffic monitoring and vehicle management.
• Transfer learning approaches provide a robust solution for complex visual recognition tasks in the automotive domain.