Deep learning based System for automatic motorcycle license plates detection and recognition

• Published in: Signal, image and video processing Vol. 18; no. 12; pp. 8869 - 8879
• Main Authors: Fathi, Abdolhossein, Moradi, Babak, Zarei, Iman, Shirbandi, Afshin
• Format: Journal Article
• Language: English
• Published: London Springer London 01.12.2024; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Computer Imaging; Computer Science; Datasets; Deep learning; Image Processing and Computer Vision; Intelligent transportation systems; License plates; Licenses; Machine learning; Motorcycles; Multimedia Information Systems; Nighttime construction; Original Paper; Pattern Recognition and Graphics; Payment systems; Recognition; Signal,Image and Speech Processing; Source code; Traffic control; Vision; Iranian; Motorcycle; Faster RCNN; SSD; License plate; YOLOv8
• ISSN: 1863-1703; 1863-1711
• DOI: 10.1007/s11760-024-03514-5

Nowadays, by increased the utilization of motorcycle the detection and recognition of its license plate play a very important role in intelligent transportation systems (ITS). ITS can be used for traffic control, violation monitoring, e-payment systems in the toll pay and parking. Several algorithms have been developed for this task and each of them has advantages and disadvantages under different circumstances and situations. By emerging deep learning based methods, they were employed to tackle the issue of automatic license plate detection and recognition. Since the deep learning models need a high volume of data for efficient training, and also each country has its license plate template, at first, it is crucial to collect proper dataset and then trains an efficient model on it. To this end, this research collected and introduced a new dataset, and then, designed a deep learning-based system for automatically detecting and identifying Iranian motorcycle license plates. At first, images that have different dimensions, angles, levels of lighting (daytime and nighttime images), were collected from various cities. Then two datasets for detection and identification are annotated and constructed from these images. Finally for implementing an efficient deep learning-based system, three networks YOLOv8, SSD, and Faster RCNN were investigated for detection and identification of license plates. The obtained results showed that the YOLOv8 network has the best result with 98.5% accuracy in the detection stage and 99% accuracy in the identification stage. The proposed YOLOv8 model was compared with deep learning-based methods and showed better performance on the collected dataset. The collected dataset and the source code of the investigated models are publicly available.