Real-Time Facemask Detection Using Deep Convolutional Neural Network-Based Transfer Learning

In response to the rapid outbreak of COVID-19, this chapter presents an automated facemask detection technology in a real-time video stream using a deep convolutional neural network based transfer learning mechanism. In the training phase, visual geometry group network (VGG16) classifier is fine-tuned using a limited medical facemask dataset with weights sharing from a large pre-trained ImageNet weight model. To build an accurate model, VGG16 needs an abundance of data. Thus, we deploy a pre-trained model to be built on ImageNet, a large visual database, for fine-tuning the VGG16 model using knowledge transfer. In the testing phase, models are tested with offline images and video streams for classification performance measurements. To test the effectiveness of the proposed method on the real-time video stream, we set up the video acquisition using the webcam and pass each frame through the XML classifiers.