Computer Vision System for License Plate Recognition and Reading

Abstract

This research paper presents the development of an automatic license plate recognition (ANPR) system based on computer vision and deep learning. The system uses the YOLOv8 model in segmentation mode, trained on a Bolivian license plate dataset, to detect and extract license plates from frames captured by an IP camera. Image processing with OpenCV is then applied to correct license plate orientation and perspective, improving character reading accuracy. For optical character recognition (OCR), the Microsoft TrOCR model is implemented, demonstrating high efficiency even with low-resolution images and without intensive preprocessing. The system was developed and tested on a laptop with modest hardware specifications, showing promising results that could be improved with greater computational power. The results indicate a significant improvement in license plate detection and recognition compared to traditional methods.