YOLO-based Automatic Number Plate Recognition with Unauthorized Font Detection

Illegal fonts used on motor vehicle number plates pose serious threats to automatic vehicle identification and law enforcement. Conventional Automatic Number Plate Recognition (ANPR) systems, trained with standard fonts, will misinterpret characters when given stylized or decorative fonts. This may lead to mistakes, like interpreting "4AN∘0088" as "LANP0088," rendering traffic monitoring systems inaccurate. This project suggests a system capable of recognizing and authenticating the illicit use of fonts on number plates. It integrates Vehicle Make and Model Recognition (VMMR), ANPR, and authentication of font styles into one deep learning-based system. MobileNet-V2 and YOLOx are utilized for detection of vehicles, and YOLOv8, EasyOCR, and SVTR-tiny are utilized for ANPR. Authentication of font styles is carried out with Convolutional Neural Networks (CNNs). Besides, Grad-CAM visualizes and describes the prediction made by the model to promote transparency and effectiveness in reclassification. This computerized tool minimizes manual verification, provides font compliance with regulations, improves road safety, and ensures more efficient enforcement of traffic rules.