Year: 2025
Type: article
Abstract: This paper examines modern methods for detecting QR codes using deep neural networks, analyzing their advantages and limitations. The study reveals that, compared to other matrix code recognition techniques, neural network-based approaches significantly enhance the speed and accuracy of both detection and decoding, while maintaining real-time performance on mobile devices. A specialized dataset of annotated QR code images was created to train and test the models. The study proposes a modification to the YOLO model, adapted for the specific task of QR code recognition through the identification of key points. Rather than traditional bounding box detection, the model focuses on recognizing a set of key points that form the QR code structure, allowing for more precise localization and facilitating the decoding process. The proposed approach accelerates the QR code localization and decoding process, ensuring a high recognition accuracy. Different configurations of the YOLOv8 neural network were examined, resulting in a QR code recognition model integrated into a mobile application. This solution demonstrated promising results, achieving efficient QR code recognition in real-world scenarios, even under challenging conditions such as poor lighting and image distortion. The model's practical implications extend to developing mobile applications capable of real-time QR code detection, particularly valuable for digital payment systems and contactless access solutions. Through thorough experimentation, the adapted YOLOv8 model consistently outperformed traditional methods in terms of both speed and accuracy, making it a viable solution for mobile-based QR code recognition. Future work will explore further refinements to the YOLO architecture and expand the dataset to enhance robustness across diverse conditions and mobile device capabilities.
Authors: ДЕНИС БРАТАСЮК, ДМИТРО ФЕДАСЮК
Institution: Lviv Polytechnic National University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4409511853