Institute of Sustainable Energy, Universiti Tenaga Nasional, Kajang 43000, Malaysia.
Communication Engineering Department, Iraq University College, Basra, Iraq.
Comput Intell Neurosci. 2021 Nov 6;2021:3971834. doi: 10.1155/2021/3971834. eCollection 2021.
Recognizing vehicle plate numbers is a key step towards implementing the legislation on traffic and reducing the number of daily traffic accidents. Although machine learning has advanced considerably, the recognition of license plates remains an obstacle, particularly in countries whose plate numbers are written in different languages or blended with Latin alphabets. This paper introduces a recognition system for Arabic and Latin alphabet license plates using a deep-learning-based approach in conjugation with data collected from two specific countries: Iraq and Malaysia. The system under study is proposed to detect, segment, and recognize vehicle plate numbers. Moreover, Iraqi and Malaysian plates were used to compare these processes. A total of 404 Iraqi images and 681 Malaysian images were tested and used for the proposed techniques. The evaluation took place under various atmospheric environments, including fog, different contrasts, dirt, different colours, and distortion problems. The proposed approach showed an average recognition rate of 85.56% and 88.86% on Iraqi and Malaysian datasets, respectively. Thus, this evidences that the deep-learning-based method outperforms other state-of-the-art methods as it can successfully detect plate numbers regardless of the deterioration level of image quality.
识别车辆牌照号码是实施交通法规和减少日常交通事故数量的关键步骤。尽管机器学习已经取得了很大的进展，但牌照识别仍然是一个障碍，特别是在那些牌照号码用不同语言书写或混合拉丁字母的国家。本文提出了一种使用基于深度学习的方法结合从两个特定国家（伊拉克和马来西亚）收集的数据来识别阿拉伯语和拉丁字母牌照的识别系统。所研究的系统旨在检测、分割和识别车辆牌照号码。此外，还使用伊拉克和马来西亚的车牌来比较这些过程。总共测试和使用了 404 张伊拉克图像和 681 张马来西亚图像。评估是在各种大气环境下进行的，包括雾、不同对比度、污垢、不同颜色和变形问题。该方法在伊拉克和马来西亚数据集上的平均识别率分别为 85.56%和 88.86%。因此，这表明基于深度学习的方法优于其他最先进的方法，因为它可以成功地检测牌照号码，而不管图像质量恶化的程度如何。
