Mask Detection App Uses Haar Cascade and Convolutional Neural Network to Alert Comply with Health Protocols

Abstract

This study aims to identify the face of a person whether wearing a mask or not wearing a mask accompanied by an appeal to the importance of wearing a mask. The contribution of this paper to science is to provide an overview of the results of accuracy, precision, recall used by the method used with data that can be accessed by many people, so that it can be developed further or can be compared. This system uses two techniques, namely the classification of whether a person is wearing a mask or not using the Convolutional Neural Network (CNN) model. The architecture used is DenseNet-12 to detect human face objects. The data used has a total of 2332 data sets, 200 of which were retrieved manually as research objects, and the rest were obtained from Kaggle. All data is evaluated using the camera in real-time. The test results show that testing scenario one has the highest score with an accuracy of 85% while testing scenario two gets results of 80%, the precision value in testing scenario one gets results of 75%, and testing scenario two has results of 88%. Scenarios 1 and 2 also have the same recall value of 100%. Based on the data analysis, it can be concluded that the use of the Haar Cascade approach and the Convolutional Neural Network with the DenseNet-121 architecture produces good performance in the case of real-time detection of masked and non-masked facial objects.