Sistema de Reconocimiento Automático de Placas Vehiculares mediante Aprendizaje Profundo para Control de Acceso en Conjuntos Residenciales

Abstract

Abstract: This document describes the techniques and models used for license plate detection through convolutional neural networks and Deep Learning, such as YOLO, CNN, CRNN, and optical character recognition (OCR) techniques including PaddleOCR, EasyOCR, and Tesseract OCR. This study was based on the use of the YOLOv11 model with multiple training datasets, which were collected under different environmental conditions, such as daytime and nighttime scenes, plates with varying degrees of rotation, and different camera angles in which they were captured on video. As support for the OCR stage, several image processing techniques were applied, allowing the recognition score of each plate to increase significantly. This approach enabled to evaluate the best trained model and test it as a prototype in a residential complex, obtaining the desired results and demonstrating that this solution can already be used in a real production environment.