Development of a Deep Learning Based Vehicle License Plate Scheme

This research developed a license plate recognition system designed for a web-based application. The developed scheme used high quality images of Nigerian license plate obtained from Google search engine and a developed dataset of vehicle license plates from the Faculty of Engineering, Ekiti State University. Rotation, exposure adjustment, blurring, and angular variation were applied on the acquired image to reduce the cost inquired in using the original image. Each image was resized to a uniform resolution of 640 x 640 pixels, standardizing input dimensions for the model. A YOLOv12 model was selected for its real-time object detection capabilities. The pre-processed and augmented images were used to train the model. The model was trained to accurately detect and localize Nigerian license plates within images, with bounding boxes drawn around identified regions. With the dataset the model achieved a precision rate of 94%, recall rate of 85.5%, and recognition accuracy of 98.08. The results for the developed model have a F1 score of 89.5% and a Mean Average Precision (mAP) of 90.2. The YOLOv12 model demonstrated strong performance in real-world scenarios, accurately predicting license plates in 94% of cases (precision) and successfully detecting 85.5% of all license plates (recall). This indicates that the model is highly reliable for practical applications, such as automated toll collection, traffic monitoring, and security systems.