Facemask Detection using the YOLO-v5 Algorithm: Assessing Dataset Variation and R esolutions

Abstract

The Covid-19 pandemic has made it imperative to prioritize health standards in companies and public areas with a large number of people. Typically, officers oversee the usage of masks in public spaces; however, computer vision can be employed to facilitate this process. This study focuses on the detection of facemask usage utilizing the YOLO-v5 algorithm across various datasets and resolutions. Three datasets were employed: the face with mask dataset (M dataset), the synthetic dataset (S dataset), and the combined dataset (G dataset), with image resolutions of 320 pixels and 640 pixels, respectively. The objective of this study is to assess the accuracy of the YOLO-v5 algorithm in detecting whether an individual is wearing a mask or not. In addition, the algorithm was tested on a dataset comprising individuals wearing masks and a synthetic dataset. The training results indicate that higher resolutions lead to longer training times, but yield excellent prediction outcomes. The system test results demonstrate that face image detection using the YOLO-v5 method performs exceptionally well at a resolution of 640 pixels, achieving a detection rate of 99.2 percent for the G dataset, 98.5 percent for the S dataset, and 98.9 percent for the M dataset. These test results provide evidence that the YOLO-v5 algorithm is highly recommended for accurate detection of facemask usage.