| Abstract |
The world has faced up with a significant healthcare challenge due to the COVID-19 pandemic. The main, convenient, and effective solution is wearing a certified mask, which can prevent approximately 80% of all respiratory infections. Therefore, many COVID-19 monitoring systems based on face mask detection have been proposed or commercially developed to provide effective supervision for public areas. This paper proposes a comprehensive in-browser face mask detection with stand-alone (server-less) and client-server architectures. They can be integrated into available real-world scenarios at public area entrances and CCTV surveillance systems. To find the best predictive face mask detection model to deploy on devices for real-time in-browser applications, we build four YOLO iterations, namely YOLOv4, YOLOv5 small, YOLOv5 nano, and YOLOX with a network size of 640 × 640. We consider CPU and GPU device deployment techniques to optimize the inference speed (FPS) with high accuracy. Experimental results present that deploying YOLOv5 small on the client-server and implementing YOLOv5 nano on the stand-alone satisfied our study’s goal to balance between adequate accuracy and real-time detection speed. This works achieves accuracy mAP of 90.8% and 29.59 FPS on GPU with client-server architecture, and mAP of 89.40%, and speed of 33 FPS on stand-alone low computing CPU resource. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2023. |
