ABSTRACT
Wearing a facemask is one of the main ways to prevent the spread of respiratory diseases such as Covid-19, so it is helpful to monitor people's facemask-wearing status through vision-based systems. In this paper, a system has been developed that divides people's face mask-wearing conditions using image processing into three classes: without a facemask, correct facemask-wearing, and incorrect facemask-wearing. For this purpose, the SSD-MobileNetV2 neural network has been used, and several hyperparameter sets have been compared for the best possible accuracy. A lightweight custom convolutional neural networks (CNN) has also been used as the second stage to improve the classification accuracy, so this stage can be used in cases where higher accuracy is required. The proposed neural network was implemented on a Raspberry-Pi3, and this system can control an entrance gate using a servo motor automatically. © 2022 IEEE.