With many applications in security, surveillance, and intelligent transportation systems, Automatic License Plate Recognition (ALPR) is a fundamental and important task. In the recent era with the growth of Deep Learning (DL) techniques, ALPR also has shown dramatically accurate results as other computer vision problems. Most of the practical ALPR systems are country-specific, for example, a lot of work has been done for Chinese, Indian, Brazilian, American, or European vehicle plates that have specific vehicle plate format. This work focuses on vehicle plate recognition for multiple formats implemented in the same country or state. For this purpose, Pakistan has been chosen where license plates have unique formatting styles even within different provinces that enhance the complexity of character segmentation and recognition, and ultimately make it a challenging problem. Moreover, due to commercialization and privacy concerns, there was no publicly available dataset of Pakistani license plates, so we have built our own dataset under different environmental conditions like sunny, foggy, or night-time. Also, the dataset contains some blurred, angled, and distorted images. We achieved 90.13% segmentation accuracy for standard number plates. After this, for correct recognition of characters, we have developed an eight layers Deep Neural Network (DNN) model and achieved 97.19% classification accuracy. To the best of our knowledge, it is the first-ever system developed for Pakistani vehicle plates using DNN where distorted images have also been considered in the dataset and a reasonable accuracy is achieved.
Paper
Full text
Pakistani Standard Vehicle Plates Recognition using Deep Neural Networks
Semantic Scholar · Computer Science · 2021
Abstract
With many applications in security, surveillance, and intelligent transportation systems, Automatic License Plate Recognition (ALPR) is a fundamental and important task. In the recent era with the growth of Deep Learning (DL) techniques, ALPR also has shown dramatically accurate results as other computer vision problems. Most of the practical ALPR systems are country-specific, for example, a lot of work has been done for Chinese, Indian, Brazilian, American, or European vehicle plates that have specific vehicle plate format. This work focuses on vehicle plate recognition for multiple formats implemented in the same country or state. For this purpose, Pakistan has been chosen where license plates have unique formatting styles even within different provinces that enhance the complexity of character segmentation and recognition, and ultimately make it a challenging problem. Moreover, due to commercialization and privacy concerns, there was no publicly available dataset of Pakistani license plates, so we have built our own dataset under different environmental conditions like sunny, foggy, or night-time. Also, the dataset contains some blurred, angled, and distorted images. We achieved 90.13% segmentation accuracy for standard number plates. After this, for correct recognition of characters, we have developed an eight layers Deep Neural Network (DNN) model and achieved 97.19% classification accuracy. To the best of our knowledge, it is the first-ever system developed for Pakistani vehicle plates using DNN where distorted images have also been considered in the dataset and a reasonable accuracy is achieved.