Comparative Study of CNN and YOLOv3 in Public Health Face Mask Detection

Coronavirus Disease (COVID-19) is gaining special concern from entire world population. The transmission of the COVID-19 virus is spreading almost in whole the world, including Indonesia which undergoing a crisis, especially in the health and economic sector. In prevention, the government is implementing Large-Scale Social Restrictions which public services or public places require people to wear masks. During this time, the detection of masks is done manually with observations from security personnel, which is time consuming. This study will apply a mask detection system (Face Mask Detection) using deep learning image processing. This study apply the most popular deep learning model which consist Convolutional Neural Networks (CNN) and You Only Look Once (YOLOv3) method. In training step, the datasets taken vary with images of faces that using head attribute such as hijabs, hats, and not using attributes. In addition, the images were taken from various countries such as Asia including Indonesia mostly, Europe, and the Americas. The system used a combination of object detection classification, image, and object tracking to develop a system that detects using a mask or not using a mask faces in images or camera videos. From the comparative analysis which developed in training and deploying step with image and camera video stream, YOLOv3 can detect accurately and faster with 4.8 FPS than CNN.