Comparative analysis of Face Mask Detection models using deep learning techniques

The spread of and the casualties due to covid-19 virus can be kept under control with human efforts and self-discipline such as sanitization and wearing masks. However, the suggested protocols could not be followed all the time due to various reasons, such as negligence or taking the protocols lightly. Hence, in this paper, a face mask classification (FMC) and detection model are being proposed that can easily differentiate between humans wearing masks and with no mask on their face. We address this problem with object detection approach using deep learning techniques. Two main classifiers used in the paper are MobileNetV2 [1] and ResNet V2 152 [2]. The detection step has been performed with Haar-cascade, Multi-Task Cascaded Convolutional Neural Network (MTCNN) and Single shot detector (SSD) [3]. The hybrid model has been compared based on accuracy, frames per second (FPS), and F1-score. MobileNetV2 and Resnet V2 152 has detected human face masks at an accuracy of 98% and 96 % respectively based on the pre-trained data. The proposed method has a great potential for being implemented in hardware as an application and establish covid-19 safety guidelines in public places to prevent/reduce the spread of the virus.