Desenvolvimento de um sistema embarcado de reconhecimento automático de placas automotivas para gerenciamento de tráfego

Abstract

Automated License Plate Recognition (ALPR) systems consist of identifying and re- cognizing text on license plates, typically from images. The potential applications for such systems include traffic enforcement, access control to private areas, and monitoring. While this type of system already exists commercially, its costs are often high due to the need for high-performance hardware to handle digital image processing, in addition to being developed with the aim of making a profit This work explores the feasibility of using low-cost embedded hardware to fulfill the purpose of an ALPR system, employing a TensorFlow Lite neural network model specifically trained for automotive plate identi- fication in images and OCR techniques, both embedded using the Tesseract OCR library and cloud services using Vision AI. Auxiliary algorithms were also developed to improve the overall system accuracy by correcting potential errors arising from the OCR stage. Additionally, an interface was built using Custom TKinter to enhance the visualization of the steps and facilitate system configurations. The results obtained for a set of 30 selected images from an image database are 93.25% accuracy for the system operating in embedded mode and 99.59% for the system operating with cloud-based OCR service.