A High-Accuracy Deformable Model for Human Face Mask Detection

Human face mask detection leverages computer vision technology to discern whether individuals in images or videos are wearing masks. Ensuring proper mask usage is crucial in settings such as hospital operating rooms and flu clinics. As deep learning advances, face mask detection has emerged as a significant research area within the computer vision field. In this paper, we propose a deformable state-of-the-art (DSOTA) model based on Deformable ConvNets v2 (DCNv2) and YOLOv8 (i.e., You Only Look Once). We use this new model to improve the accuracy of face mask detection. Our experimental results show that the integration of DCNv2 and YOLOv8 significantly improves the accuracy of face mask detection. The average highest accuracy rate of the YOLOv8n model is 91.7%, and the average highest accuracy rate of the DSOTAn model is 94.4%. The average highest accuracy rate of the YOLOv8s model is 97.0%, and the average highest accuracy rate of the DSOTAs model is 97.4%. These promising results underscore the potential of our approach for practical applications and further exploration in the computer vision domain.