Evaluation and Classification of Face Shield Use with Convolutional Neural Networks

To prevent the spread of the coronavirus, different protocols and security rules were implemented to allow for the gradual return of activities in certain areas, one of these protocols is the mandatory use of face shields to enter establishments where there is a high flow of people. In this context, it is necessary to have systems that help to control the correct use of the face shield, techniques such as Convolutional Neural Networks (CNN) can be used to implement these systems through the use of supervised learning applied to images obtaining a great performance in classification tasks. This work focuses mainly on the training and deployment of a CNN capable of identifying the correct use of the face shield. Two approaches were tested: training the CNN from scratch and using the transfer learning technique using as input data images collected in a real scenario as well as freely available repositories. The results obtained in terms of accuracy showed superior performance for the model trained from scratch compared to the model trained using the transfer learning technique.