Detection and Recognition of Bangla Nameplates Using YOLOV6 and BLPNET Models

Abstract

With an emphasis on Bangladeshi car number plates, this thesis offers a successful method for identifying and recognizing license plates. Character recognition, plate detection, and dataset creation are the three stages of the study procedure. A collec tion of 1,000 live-captured and online source pictures of Bangladeshi license plates was assembled from internet resources. With the help of the hybrid model BLPNET, which combines VGG19 and RESNET50, and the YOLOV6 object detection method, we were able to obtain an F1 score of 96.4% for character recognition and a pre cision of 96% for license plate detection. The system addresses challenges unique to Bangladeshi number plates, such as complex design, diverse weather conditions, and poor image quality due to occlusion in dense urban environments. The study contributes to the development of smart traffic management systems, aligning with Bangladesh’s vision of becoming a "smart" nation. This work also highlights the gap in public datasets for Bangla license plates, presenting a new dataset and models that push the boundaries of license plate recognition technology.