Mask-Vision: A Machine Vision-Based Inference System of Face Mask Detection for Monitoring Health Protocol Safety

Abstract

To avoid adversely affecting community health and the global economy, effective ways to limit the COVID-19 pandemic require constant attention. In the absence of efficient antivirals and insufficient medical resources, WHO recommends several methods to minimize infection rates and prevent depletion of scarce healthcare resources. One of the non-pharmaceutical treatments that can be used to decrease the primary source of SARS-CoV2 droplets expelled by an infected individual is to wear a mask. Irrespective of disagreements about medical resources and mask types, all governments enforce the wearing of masks that cover the nose and mouth by the general population. In the next years, the suggested mask detection models might be a valuable tool for ensuring that safety measures are followed correctly. The YOLOv3 model, a deep transfer learning object identification state-of-the-art approach, is used to create a mask detection model in this research article. The suggested model's exceptional performance makes it ideal for video surveillance equipment. The suggested approach focuses on creating an enhanced dataset from a 300-image dataset utilizing data augmentation techniques such as image filtering. The Data augmentation-based mask detection model's mean average precision was found to be 89.8% during training and 100% during overall testing, with detection per frame accuracy ranging from 40.03% to 65.03%.