Abstract
In addition to rapid vaccination, predicting possible trajectories of the COVID-19 pandemic is critical to health-care-related policy decisions and infrastructure planning. Growing evidence shows that face masks and social distancing can considerably reduce the spread of respiratory viruses like COVID-19. However, the current pandemic trajectory predictions take overly simplified policy input rather than actual observations of face masks and social distancing practices in a crowd. Thus, it is crucial to monitor and understand the extent of masking practices and assess the safety level in a scalable manner.
This paper proposes a novel face masking detection system for Modeling Safety Index in Crowd (Mosaic), a Machine Learning (ML)-based approach for detecting masking in a crowd by building new dense mode crowd mask datasets. Mosaic detects, counts, and classifies the crowd's masking condition and calculates spatiotemporal Safety Index (SI) values for each community instead of detecting individual masking cases. SI data can be shared or published to calculate the area-based SI maps (as opt-in data) for assisting effective policy decisions and relief plans against COVID-19. The experimental results show that Mosaic detects various conditions and types of masking states and calculates SI values of a crowd effectively.