Enhancing campus security and vehicle management with real-time mobile license plate reader app utilizing a lightweight integration model

Ahmad Radzi, Syafeeza and Kamarozaman, Muhammad Haziq and Wong, Yan Chiew and Abdul Hamid, Norihan and Mohd Saad, Wira Hidayat and Abdul Samad, Airuz Sazura (2024) Enhancing campus security and vehicle management with real-time mobile license plate reader app utilizing a lightweight integration model. Journal Of Engineering Science And Technology, 19 (5). pp. 1672-1692. ISSN 1823-4690

Text 19_5_07.pdf Restricted to Registered users only Download (1MB)

Abstract

The increasing number of vehicles owned by campus residents, combined with a limited number of staff parking lots, poses challenges for security personnel in distinguishing between staff, student vehicles, and visitors. Additionally, the presence of untracked external visitors and the potential manipulation of vehicle registrations by residents pose safety risks. Implementing a License Plate Detection and Recognition (LPDR) mobile app could ease the burden on security patrols. However, traditional LPDR systems face real-world limitations, including various backgrounds, illumination, weather, and distances. Therefore, opting for a deep learning-based LPDR approach is the way forward. Nevertheless, implementing deep learning on resource-constrained mobile devices demands significant storage and computational power. This may lead to user hesitation when it comes to downloading the app onto their mobile devices. This paper introduces an automated Android mobile app for real-time license plate reading, integrating a lightweight YOLOv8n for license plate detection and ML Kit Optical Character Recognition (OCR) for text recognition. The inference integration model is performed on the mobile device to streamline security tasks, aiming to assist security personnel in identifying any wrongdoing by campus residents and visitors. The lightweight model addresses mobile device resource limitations by implementing an on-device machine learning model and storing vehicle ownership information on a cloud server without compromising accuracy. The plate detection accuracy is approximately 97.5%, the character recognition accuracy is around 91.2% which is on par with other LPDR existing works. The study results in a precise end-to-end mobile app for license plate reader, benefiting patrol units with low error rates and real-time capabilities. The mobility feature enables swift security responses, ultimately enhancing overall campus safety.

Item Type:	Article
Uncontrolled Keywords:	Android, Campus safety, End-to-end mobile app, Object detection, Real-time detection, Real-world scenarios.
Divisions:	Faculty Of Electronics And Computer Technology And Engineering
Depositing User:	Norfaradilla Idayu Ab. Ghafar
Date Deposited:	06 Jan 2025 11:15
Last Modified:	06 Jan 2025 11:15
URI:	http://eprints.utem.edu.my/id/eprint/28168
Statistic Details:	View Download Statistic

Actions (login required)

View Item