On the use of Deep Learning Enabled Face Mask Detection For Access/Egress Control Using TensorFlow Lite Based Edge Deployment on a Raspberry Pi
Reenu Mohandas, Mangolika Bhattacharya, Mihai Penica, Karl Van Camp, Martin J. Hayes · 2021
This paper presents a face mask detection system that can be easily retrofitted to a standalone Access and Egress control system and can be deployed using any Edge processing subsystem. In this work, the widely available Raspberry Pi is used incorporating novel TensorFlow lite based training methods to demonstrate the utility of the approach. This subsystem achieves real-time face mask detection with high accuracy of over 89% in face mask detection at a confidence level of greater than 90% with excellent detection speeds of less than 3ms. The performance of the Raspberry Pi based implementation compares very favorably with GPU desktop-based alternative approaches that require excellent real time cloud enabled communications. The system is shown to work robustly with minimal human intervention and can be easily deployed at entrance points to lab or other employee facilities where face masks are mandatory for employee safety. The proposed algorithm consists of a single stage object detector together with an adaptive training dataset. This dataset comprises a hybrid combination of locally collected training data together with publicly available datasets and it is shown to be most appropriate for the face mask detection use case at hand.