YOLO Based Real-time Bangladeshi Vehicle License Plate Identification

Abstract

(ALPR) Automatic License Plate identification plays a crucial role in traffic management, security applications, tools collection and various aspects of modern life. Every country has their own unique license plate model. For this reason, in every country, we need to use different dataset containing their license plate and different models to develop license plate identification. This research explores the application of YOLOv9, which are a powerful deep learning model, for real-time Identification of Bangladeshi license plates and it is so fast and efficient. The proposed approach involves training YOLOv9 on a comprehensive dataset, which has a large number of Bangladeshi license plates images of various angle, various lighting condition and various weather condition taken by mobile phone camera by me. This proposed system addresses the specific challenges associated with Bangladeshi license plates, including their unique format, classes, types and potential variations in lighting and environmental conditions. As of my knowledge there are no robust dataset available for this research. I have collected more than 4000 Bangladeshi license plate images with various angles, weather conditions and different lighting conditions. Importantly, I have collected all the classes and types of Bangladeshi license plate images for robust recognition. I train the dataset with YOLOv8m & YOLOv9c framework. YOLO (You Only Look Once) is known for fastest and efficient recognition of small characters like license plate. My proposed model achieved 96.54% accuracy for detecting all types of license plate in Bangladesh including Army, Navy, Air-force. There are existing papers available there has up to 95% accuracy rate for detecting license plate although they did not use all the classes and types of images of Bangladeshi license plates. So, I used all the types and classes of images for robust detection of Bangladeshi license plate. The proposed solution improves ALPR by including YOLOv9, a cutting-edge deep learning model, to enable rapid and accurate license plate identification. For recognition characters I used EasyOCR technique and I achieved 86.67% accuracy of identification the characters. My proposed approach has the potential to transform traffic management, security, and applications in Bangladesh, allowing for automated toll collection, improved traffic analysis, and enhanced security measures.