Título: Investigating image pre -processing techniques to enhance Brazilian license plate recognition
Autores: Michel Lutegar, André Coelho, Marceu Filho, Rigel Procópio Fernandes, Thiago Silva de Souza & Clayton J. A. Silva
Resumo: This paper presents the development and optimization of an embedded real-time license plate recognition system designed with applications to public security operations in Brazil. The system aims to equip police vehicles and inspection posts with instant vehicle identification capabilities to detect stolen, cloned, or flagged vehicles during routine patrols and checkpoints. Vehicle-related crimes, including theft and cloning, pose significant challenges to public safety in Brazil, requiring law enforcement agencies to rapidly identify suspicious vehicles from extensive databases. Traditional manual verification methods are time-consuming and inefficient for real-time operations. Our proposed solution integrates YOLO detection algorithms with OCR technology in an embedded system capable of real-time processing and automatic database cross-referencing. To optimize system performance, we evaluate eight different image preprocessing methods. Each preprocessing approach is tested with threshold values ranging from 0.0 to 0.8 using a dataset of 65 Brazilian license plates to determine optimal parameter combinations for maximum OCR accuracy. Experimental results demonstrate significant performance variations across preprocessing methods, with the Resized2x approach achieving 66.7% accuracy at threshold 0.8 and Inverted preprocessing with 0.4 threshold reaching 11.8% accuracy in optimal configurations. The system architecture enables instant image capture through vehicle-mounted cameras, real-time plate processing, and immediate database queries within seconds of detection. These findings contribute to the development of an effective autonomous surveillance system that can substantially improve police operational efficiency and citizen safety through rapid identification of vehicles of interest during active patrol operations.
Palavras-chave: license plate recognition; OCR; image preprocessing; YOLO; Brazilian traffic systems.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1191901
Artigo em PDF: CBIC_2025_paper1191901.pdf
Arquivo BibTeX:
CBIC_2025_1191901.bib