An Automated Approach for the Recognition of Bengali License Plates

TL;DR

This paper presents a hybrid automated system for recognizing Bangladeshi vehicle license plates, combining YOLO for detection, Otsu's Thresholding for segmentation, and CNN for character recognition, achieving 81% detection accuracy.

Contribution

It introduces a novel hybrid approach specifically tailored for Bangladeshi license plates, integrating multiple techniques for improved recognition accuracy.

Findings

81% correct prediction in license plate detection

Effective segmentation using Otsu's Thresholding

Successful application of CNN for character recognition

Abstract

Automatic Number Plate Recognition (ALPR) is a system for automatically identifying the license plates of any vehicle. This process is important for tracking, ticketing, and any billing system, among other things. With the use of information and communication technology (ICT), all systems are being automated, including the vehicle tracking system. This study proposes a hybrid method for detecting license plates using characters from them. Our captured image information was used for the recognition procedure in Bangladeshi vehicles, which is the topic of this study. Here, for license plate detection, the YOLO model was used where 81% was correctly predicted. And then, for license plate segmentation, Otsu's Thresholding was used and eventually, for character recognition, the CNN model was applied. This model will allow the vehicle's automated license plate detection system to avoid any misuse.