Abstract

Over the past decade, computer vision has emerged as a pivotal field, focusing on automating systems through the interpretation of images and video frames. In response to the global impact of the COVID-19 pandemic, there has been a notable shift towards utilizing computer vision for face mask detection. Face masks, endorsed by international health authorities, play a crucial role in preventing viral transmission, prompting the development of automated monitoring systems in various public settings. However, existing artificial intelligence (AI) technologies' effectiveness diminishes in congested environments. To address this challenge, the study employs a meticulously fine-tuned YOLOv4 model for identifying instances of mask non-compliance in accordance with COVID-19 Standard Operating Procedures (SOPs). A distinctive feature of the training dataset is its inclusion of images featuring Muslim women with both half and full-face veils, considered compliant with face mask guidelines. The dataset, comprising 5800 images, including veil images from various sources, facilitated the training process, achieving a comparatively good 97.07% validation accuracy using transfer learning. The adaptations, coupled with a custom dataset featuring crowded images and advanced pre-processing techniques, enhance the model's generalization across diverse scenarios. This research significantly contributes to advancing computer vision applications, particularly in enforcing COVID-19 safety measures within public spaces. The tailored approach, involving model adjustments, underscores the adaptability of computer vision in addressing specific challenges, highlighting its potential for broader societal applications beyond the current global health crisis.