An End-to-End Automated License Plate Recognition System Using YOLO Based Vehicle and License Plate Detection with...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Transportation Systems

Background:

• Existing Automatic License Plate Recognition (ALPR) methods struggle with generalization in real-world conditions due to reliance on prior knowledge and fixed processing rules.
• Intelligent Transportation Systems (ITS) require robust ALPR for effective surveillance and traffic management.

Purpose of the Study:

• To develop and evaluate an end-to-end, generic ALPR pipeline using YOLO-based models.
• To improve license plate (LP) detection and recognition accuracy without requiring prior knowledge or additional inference steps.

Main Methods:

• A YOLO-based pipeline integrating vehicle detection (VD), license plate (LP) detection, and recognition.
• Utilized YOLO v2 for initial vehicle detection and YOLO v4 for subsequent stages, incorporating extensive data augmentation.
• Included a vehicle classifier for emergency vehicles and heavy trucks within the pipeline.

Main Results:

• Achieved an average license plate recognition accuracy of 90.3% across five public datasets from different regions.
• Demonstrated competitive LP recognition accuracy compared to current state-of-the-art methods.
• Maintained acceptable frames per second (FPS) performance on a low-end GPU.

Conclusions:

• The proposed YOLO-based end-to-end ALPR pipeline offers a robust and versatile solution for ITS.
• The generic approach overcomes limitations of traditional ALPR methods, showing strong generalization capabilities.
• The system provides high accuracy and efficiency, suitable for real-world surveillance applications.