Face Mask and Face Shield Detection Using Image Processing with Deep Learning and Thermal Scanning for Logging System

With the crisis of the COVID-19 pandemic, it has become apparent in the Philippines that protocols need to be put in place that ensures the health and safety of the people. Included in those protocols is contact tracing and the proper use of masks and shields. The purpose of this research is to develop a system of face mask and face shield detection using image processing with deep learning and thermal scanning for logging system to automate the task of surveying and compliance to wearing of mask and shield. A model for classifying the five classes: face mask, face shield, face mask and face shield, none, and no face was created and trained using the MobileNet architecture, with collected dataset using the Maixduino camera. An overall accuracy of the entire system was found to be at 90%. No face and none classifications have provided results of above 90% in precision, sensitivity, specificity, and F1-Score. While values of the only mask, only shield, and both fluctuate in values in computations, their F1-scores still falls within the range of 80%-90% in performance. The implementation of the MobileNet model on the Maixduino board for was successfully accomplished with considerable classification capability.