Accurate Real-Time Face Mask Detection Framework Using YOLOv5

Abstract

After the COVID-19 pandemic, wearing a mask has become a must because it decreases the probability of infection by 68%. That is why a fast and accurate automatic mask detection is crucial to public institutions. In this paper, we present an accurate framework for real-time mask detection using YOLOv5 object detection algorithm. Our framework consists of four stages: image preprocessing by normalization and adding noise, adding negative samples and data augmentation then the detection core based on a modified version of YOLOv5. The proposed framework achieves 95.9% precision and 84.8% mean average precision using the Face Mask Detection dataset with a 10 milliseconds inference time.