ALPR System Perspective Adjustment: New Automatic License Plate Recognition Approach for Brazilian Mercosur Model Vehicle Plates

Improved detection techniques have been identified in current research. The purpose of this study is to detect license plates on the new Mercosur model using different computational methods. These methods not only correct the license plate perspectives but also aim to identify Mercosur license plates with the YOLOv5s, YOLOv5x, YOLOv7, and YOLOv7x architectures. In addition, various character recognition algorithms, including Pytesseract, Easy OCR, Keras, Terresact.js and the ResNet-50 network, will be employed for reading the characters on Mercosur model license plates post-detection. The training datasets utilized were UFC-ALPR and UFC-LPR, encompassing 2,586 license plate images of old models and 1,100 images of Mercosur models, respectively. The most effective models suggested a combination of Yolov7x for detecting license plates and Tesseract.js for optical character recognition. The two models attained a 99.10% detection accuracy and 91.07% OCR accuracy while reading characters of Mercosur plates for genuine validation, eclipsing the current state-of-the-art.