Face Mask Wearing Condition Classification using YOLOv5 trained with Hybrid Dataset

Face mask detection was not a major needed application of computer vision till 2020. After the deadly COVID-19 disease spread worldwide, several measures were taken to stop the virus from further spreading. Wearing face masks is the simplest and best precautions one can take to reduce the risk of getting infected from the virus.In crowded places like airports, malls and big companies masks are mandatory for the safety of everyone. Governments made mask compulsory in public places and had to ensure that everyone wore mask properly. Computer vision and deep learning can be used to detect people not wearing mask automatically. Many papers have discussed or proposed different models to detect face mask. Most of the models are trained to detect with and without mask. In this paper third category i.e. improper mask has been added which detects the case where mask is not worn properly. The model used for mask detection is YOLOv5 (You Only Look Once Version 5) which is the latest(still being upgraded) version of YOLO. YOLO is a type of Single Shot Detector and is very fast in object detection. This model was trained with hybrid data-set which contains 1000 images of each class that is ‘with mask’, ‘without mask’ and ‘improper/partial mask’ which makes total of 3000 images. The paper presents performance of YOLOv5 in detecting face mask when trained with the hybrid dataset.