Effective real-time face mask detection on NVIDIA edge devices

ABSTRACT

During these difficult times of COVID-19, people are struggling to return to their normal routines, including going back to schools and workspaces. To prevent the spread of the disease, wearing face masks is essential for everyone to protect themselves and the ones around them. However, challenges arise in regard to enforcement of wearing masks in large crowds such as at educational centres and public transportation. This paper proposes a robust automatic system for face mask detection using transfer learning kits from NVIDIA. Based on the backbone of Resnet-18, the model results in high accuracy in the distinguishing of persons who do and do not wear masks. Leveraged by the NVIDIA edge accelerator, the system can run in real-time environments, making it applicable in various venues. Its feasibility was demonstrated by deploying the approach in an education centre in Hong Kong. © 2022, Hong Kong Institution of Engineers. All rights reserved.