As COVID-19 continues to spread across the globe, government leaders and individuals are finding ways to slow the spread of the virus. The World Health Organization (WHO) recommended in March 2020 to wear a face-covering to prevent people from breathing out tiny droplets that may contain the virus. Properly wearing a face mask means the virus transmission can be lowered. However, watching every person in a crowded area for who is properly wearing or who is not wearing a face mask is a daunting task. This study aims to process face images in a video frame and detects if a person is wearing a mask, not properly wearing a mask (nostrils are visible), and not wearing a mask at all. Using an inception network trained with 4,789 face images with three classifications, the model was able to classify faces with 98.5% accuracy using the training images and 98.8% accuracy using the validation images. All of the training images are black and white images with 16✕16 pixels of resolution. The trained inception network has three inception modules, three max pooling layers, two fully connected layers, and a dropout layer. The frame rate of the video in which the network outputs is almost the same as the source video.
Volume 12 | 07-Special Issue