A Novel Facial Mask Detection Using Fast-YOLO Algorithm

The extremely high transmission rate of the COVID-19 has made the supply of medical resources in countries around the world in short supply. The implementation of quarantine in order to avoid group infections has a serious impact on the economy, transportation, education and other aspects. Epidemic prevention will be a routine task that needs to be carried out for a long time and cannot be neglected. In view of the fact that wearing masks is currently an effective method of epidemic prevention, and the current face detection models are not effective for masked faces, and pedestrians who have not worn masks in the correct way. It may spread the epidemic. This research will establish a face data set with three kinds of annotations, and combine a variety of deep learning convolutional neural network architectures and methods to design a face detection model that can quickly train and detect wearing a mask, not wearing a mask, and wearing a mask incorrectly faces. In the hope of contributing to the epidemic prevention, we use an adaptive algorithm to adjust the image size to reduce unnecessary operations, and modify the CIOU_LOSS error function to speed up the operation. Experiments have confirmed that our algorithm saves 70% of the time compared to YOLO v5m with the same accuracy.