Citation

Lau, Jimmy Ch Weng and Abdul Karim, Hezerul and AlDahoul, Nouar (2024) An Evaluation of Various Pre-trained Object Detection Models for Complex License Plate. In: 2024 Multimedia University Engineering Conference (MECON), 23-25 July 2024, Cyberjaya, Malaysia.

Text An Evaluation of Various Pre-trained Object Detection Models for Complex License Plate.pdf - Published Version Restricted to Repository staff only Download (491kB)

Abstract

Modern computer vision techniques for analyzing images and videos, particularly in license plate recognition (LPR), have gained significant importance across multiple domains such as highways, parking management, and law enforcement. Precise detection and recognition of license plates play a crucial role in ensuring efficiency and security standards are met at the highest levels. The datasets used in this research were obtained through collaboration with Tapway Sdn. Bhd. This dataset consists of 264 images showing blurriness, overexposure, extreme character distortion, and varying sizes of Malaysian registered license plates. Additionally, it contains 792 high-quality images of Malaysian license plates. This research employs the YOLOv8, YOLOv10, YOLOv5, and Roboflow 3.0 Object Detection (Fast) models, which are well-established object detection models widely used by developers worldwide. All models undergo training with 36 different alphabets and numbers across images of diverse quality. The research focuses on cross-dataset experiments and specific training, validation, and testing on a single dataset using three different approaches. The study includes a comparative analysis with other well-known license plate recognition techniques, particularly the pre-trained Object Character Recognition (OCR) models, such as Keras OCR. In the crossdataset approach, YOLOv5 achieved an accuracy of 83%, outperforming YOLOv8, which had 50.75% accuracy, and YOLOv10, which had 48.10% accuracy. Keras OCR recognized characters in the license plates with 40.53% accuracy.)

Item Type:	Conference or Workshop Item (Lecture)
Uncontrolled Keywords:	Optical Character Recognition. License Plate Recognition
Subjects:	Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions:	Faculty of Engineering (FOE)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	06 Feb 2025 05:08
Last Modified:	06 Feb 2025 05:08
URII:	http://shdl.mmu.edu.my/id/eprint/13361

Downloads

Edit (login required)