Detecção de placas de veículos com foco na proteção de dados pessoais

Bruno José Souza; Alessandro Zimmer; Roberto Zanetti Freire

doi:10.20906/sbai.v1i1.2639

Bruno José Souza Programa de Pós-Graduação em Engenharia de Produção e Sistemas Escola Politécnica, Pontifícia Universidade Católica do Paraná, Curitiba, Paraná
Alessandro Zimmer Center of Automotive Research on Integrated Safety Systems and Measurement Area - CARISSMA, Ingolstadt, Bayern
Roberto Zanetti Freire Programa de Pós-Graduação em Engenharia de Produção e Sistemas Escola Politécnica, Pontifícia Universidade Católica do Paraná, Curitiba, Paraná

DOI: https://doi.org/10.20906/sbai.v1i1.2639

Keywords: Vehicle license plate, Deep learning, Data protection, Object detection, YOLO

Abstract

In recent years, there has been an increase in the development of systems related to data protection. Following the recent General Data Protection Regulation (GDPR), systems are being adapted to capture images without compromising the safety of users through improper disclosure. This paper intended to evaluate the performance of different versions of deep learning algorithms aimed at the detection of vehicle license plates. The research proposes the adaptation of one method to hide the license plates that were detected in images, thus providing the protection of personal data. The selected deep learning techniques were YOLO v3 (You Only Look Once) and YOLO v4. The training of the selected algorithms was performed with the THI License Plate Dataset (TLPD). The YOLO v3 algorithm reached an Average Precision (AP) equal to 98.68%, recall of 98.00%, and Average Intersection over Union (Average IoU) of 79.38%, being elected the best among the techniques compared in this study.