ABSTRACT
Corona virus disease-2019 (COVID-2019) has impacted on many social behaviours and has put forth some cautiousness in day-to-today life. Therefore, to remove the barrier of fearful life, it is essential to monitor the preventive guidelines suggested by the world health organization. The very first guideline to be followed is to wear a mask and maintain social distance. In order to implement this in a super populous country like India, the administration used very coercive steps. To aid the administration, this paper provides a simple and easy to implement deep learning technique for the detection and recognition of COVID norm violators. Given an unconstrained/ constrained real-time video, the proposed framework uses YOLOv4 model for person localization, height-width comparison for evaluating social distance, and a customized YOLOv4 model for face mask detection. Once the proposed algorithm localizes the violators, it identifies them using convolutional neural network-based face recognition library. The evaluation metrics on benchmark datasets as well as real-time data are obtained. The proposed framework outperforms existing solutions with mAP (mAP @ 0.50 i.e. Mean Average Precision) of 0.9395 on YOLOv4. Comparison of proposed technique with the existing literature illustrates the better trade-off between accuracy and complexity. © 2022, Ismail Saritas. All rights reserved.