Real-time Face Mask Types Detection to Monitor Standard Operating Procedure Compliance Using You Only Look Once-based Framework

Abstract

Real-time face mask types detection using image processing and deep learning model had seen enormous promise in real-world applications. Due to the spread of Covid-19, the practice of wearing face masks in public areas is used to safeguard people from the virus. However, to manually detect the type of face masks used can be difficult, hence this project aims to design and develop a real-time face mask detection model that can detect types of face masks worn by an individual which include 1) surgical masks, 2) KF94, 3) N95, 4) cloth or 5) double-masking. It could also identify if an individual is wearing the face masks incorrectly. This project is developed using the modified waterfall methodology. There are four phases in the methodology: (i) Requirement Analysis, (ii) Design, (iii) Implementation, and (iv) Testing. The data used for training and testing in this project was collected from available images on the internet. The data were pre-processed to remove any unwanted images and each image is then annotated with appropriate classes. The detection model was built using the You Only Look Once version 3 (YOLOv3) framework.