A Novel IoT-Enabled System for Real-Time Face Mask Recognition Based on
Petri Nets
Abstract
Due to Coronavirus Disease 2019 (COVID-19), many countries have
formulated pandemic prevention regulations, requiring the masses to wear
a face mask before entering public places and taking public
transportations. However, if the entrances of some places are manually
checked to see whether people are wearing a face mask or not, it becomes
not only labor-intensive and time-consuming, but also inefficiently
checking each passer-by. Therefore, this paper aims to develop a face
mask recognition system based on an edge computing platform. The
traditional manual inspection control method is replaced by artificial
intelligence (AI) technology to achieve automatic recognition and
control. As an edge computing platform, Jetson Nano is an embedded
system equipped with an AI platform, which can be used for object
detection and image classification. Developed by Ultralytics LLC, a
YOLOv5 model using the PyTorch framework runs on the edge computing
platform, featuring high speed, high precision, and small size.
According to the model training results, the average precision
(AP) reaches 95.41%, while the mean average precision
(mAP) records 94.42%. The average single-class running time is
0.016 seconds, and the file size of training model is 3.8MB. The
recognition distance is up to 8m, and the maximum face rotation angle is
90°. In addition, a Petri net software tool, WoPeD, with graphical
features based on mathematical theories, is used to verify the mask
recognition system; and ensures the system has acceptable precision and
recall values.