This model uses Haar Cascade classifiers to detect Yes eyes, No nose, No mouth to ensure that mouth and nose are covered.
Traditional approach is to train a convolutional neural network on masked and non masked images. Haar classifiers are faster as compared to a CNN an don't require model training time.
The below webcam stream recording shows how the model performs :-
