Title: Segmentation-free Bangladeshi License Plate Recognition Using YOLO with Heuristic Bounding Box Refinement

Issue Number: Vol. 11, No. 1
Year of Publication: March - 2021
Page Numbers: 1-9
Authors: Mahmudul Hasan Bhuiyan, Muhammad Anwarul Azim, Mohammad Khairul Islam, Farah Jahan
Journal Name: International Journal of New Computer Architectures and their Applications (IJNCAA)
- Hong Kong

Abstract:


Automatic License Plate Recognition is the core of Intelligent Transportation System because of its diverse set of applications. There are some online APIs such as OpenALPR, Sighthound, etc. available for recognizing license plates of some countries; however, no API is available for recognizing Bangladeshi license plates. The aforementioned APIs cannot even localize Bangladeshi license plates. In this paper, we employ the Fast YOLO detector for the first time to Bangladeshi license plate recognition. We also propose an aspect ratio-based bounding box refinement technique that experimentally outperforms the conventional way of fixed padding. Various performance evaluation metrics show that our proposed system not only provides impressive performance but also outperforms the state-of-the-art methods of license plate recognition. Additionally, we also introduce an annotated dataset containing samples of different challenging situations which will help to reduce the scarcity of benchmark datasets in the domain