Robust Approach for Detecting Face Mask Using Deep Learning and Its Comparative Analysis

ABSTRACT

The world is fighting against the novel coronavirus, and a lot of people have lost their lives with the scenario getting bad to worse and worst. This infection is communicated from one individual to another while wheezing or talking as drops. To prevent Covid-19, wearing masks is very beneficial. In this paper, an existing model, ‘DenseNet201’, is being modified to efficiently track the persons who are wearing a mask or not. This system uses a convolutional neural network (CNN) and computer vision to limit the risk of Covid-19 and make sure nobody violates the rule. The dataset used in the process contains two classes, namely 'with mask' and 'without a mask'. Data pre-processing and splitting take place before the model training;then comparative analysis has been made in between the modified versions of the three transfer learning models, viz. DenseNet201, InceptionResnetV2, and ResNet101V2 to validate the modified model's efficiency. Results suggest that the revised version of DenseNet201 is very effective and can detect the events where face masks are not used at all or in an improper manner, with an accuracy of 98.90%. Various other metrics for performance are also being evaluated and reported in the paper. This model can work with images and videos/CCTV cameras using the help of OpenCV, TensorFlow, and Keras. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.