Deep Learning Based Framework for Iranian License Plate Detection and Recognition
This work addresses license plate recognition for Iranian applications, such as traffic management, but is incremental as it applies existing methods to a new dataset.
The paper tackles Iranian license plate detection and recognition by proposing a deep learning framework using YOLOv3 for detection and Faster R-CNN for character recognition, achieving high accuracy (e.g., 99.6% mAP for detection and 98.8% accuracy for recognition) and fast detection speed (23ms).
License plate recognition systems have a very important role in many applications such as toll management, parking control, and traffic management. In this paper, a framework of deep convolutional neural networks is proposed for Iranian license plate recognition. The first CNN is the YOLOv3 network that detects the Iranian license plate in the input image while the second CNN is a Faster R-CNN that recognizes and classifies the characters in the detected license plate. A dataset of Iranian license plates consisting of ill-conditioned images also developed in this paper. The YOLOv3 network achieved 99.6% mAP, 98.26% recall, 98.08% accuracy, and average detection speed is only 23ms. Also, the Faster R-CNN network trained and tested on the developed dataset and achieved 98.97% recall, 99.9% precision, and 98.8% accuracy. The proposed system can recognize the license plate in challenging situations like unwanted data on the license plate. Comparing this system with other Iranian license plate recognition systems shows that it is Faster, more accurate and also this system can work in an open environment.