Face Mask Protocol Detection Using YOLOv8

Abstract

A facemask is crucial for choosing the right type and ensuring its effectiveness in its intended use. By prioritizing proper coverage, filtration, fit, comfort, and durability, we can maximize the benefit of facemasks in protecting individuals and communities from the spread of airborne illnesses and some other dust. Facemask protocol detection has emerged as a significant technology in response to the global pandemic, aiming to curb the spread of infectious diseases. This technology has made advancements in recent years, offering promising solutions for public health and safety. This study uses machine-learning approaches to build a facemask protocol detection, and this research was to develop, test, and evaluate machine learning methods for predicting the facemask wearing, incorrect form wearing, and not wearing a facemask. To carry out the study successfully, it was essential to detect the facemask protocol. Bounding box, passing class balance the dataset, model training using YOLOv8s, YOLOv8m, and YOLOv8x model testing, and finally comparing models using evaluation metrics are also methods used to implement and design for facemask protocol detection. We used collected datasets and downloaded them from Kaggle public datasets for this research. Each balanced severed dataset class is divided into three assigning classes: face mask-wearing, incorrect form wearing, and not wearing facemasks. The experiment was conducted on three dataset types, obtaining three models for each dataset. When compared to YOLOv8s, YOLOv8m, YOLOv8x models. The YOLOv8s model produced greater test accuracy on the collected dataset, and in a public datasets, YOLOv8m will be more good accuracy. In the collected dataset, Precision Recall mAP is 0.995, F1 Confidence 0.99, and Precision Confidence 0.960, in 20 epochs. In the public datasets, Precision-Recalll mAP is 0.994, F1 Confidence 0.98, and Precision Confidence 0.936, epoch.