Iranian License Plate Recognition Using a Reliable Deep Learning Approach

Document Type : Article

Authors

Faculty of Mechanical Engineering, Tarbiat Modares University, Tehran, Iran

Abstract

The issue of Automatic License Plate Recognition (ALPR) has been a challenging one in recent years, with various factors such as weather conditions, camera angle, lighting, and different characters on license plates. However, thanks to the advances made in deep neural networks, it is now possible to use specific types of neural networks and models to recognize Iranian license plates. In the proposed method, license plate recognition is done in two steps. Firstly, the license plates are detected from the input image using the YOLOv4-tiny model, which is based on the Convolutional Neural Network (CNN). Secondly, the characters on the license plates are recognized using the Convolutional Recurrent Neural Network (CRNN) and Connectionist Temporal Classification (CTC). With no need to segment and label the characters separately, one string of numbers and letters is enough for the labels. The successful training of the models involved using 3065 images of license plates and 3364 images of license plate characters as the desired datasets. The proposed method boasts an average response time of 0.0074 seconds per image and 141 frames per second (fps) in the Darknet framework and 0.128 seconds per image in the TensorFlow framework for the License Plate Detection (LPD) part.

Keywords

Main Subjects

Scientia Iranica

Articles in Press, Accepted Manuscript
Available Online from 18 May 2024

Files

Share

How to cite

Statistics