Masked Face Recognition for Smart City Utilization

Abstract

Recent international projects to create smart cities that maintain sustainability have significantly improved humanity. Incorporating deep learning and facial recognition in intelligent cities guarantees security in public places. If facial recognition is accurate, it can hasten such activities, enhancing their convenience and adding intelligence to city life. Numerous facial recognition tasks now face significant challenges due to the increasing use of face masks. A thorough surveillance is necessary due to the widespread usage of face masks in many communities, as multiple security evaluations suggest that face masks may be used to hide identities. Consequently, using the uncovered region of the masked face for facial recognition in smart cities has become essential. In the area of intelligent city surveillance, biometrics, smart cards, law enforcement, and security information, this technology is widely used. Therefore, this work proposes a framework for developing a face recognition system for masked face images in intelligent cities. A transfer learning approach has been employed to train the Facemask Detection dataset by integrating YOLOv8 with the HOG algorithm. Furthermore, an examination of the YOLOv8 algorithm's performance compared to other algorithms has been provided. The simulated results confirm that the system is robust when identifying individuals in masked face images. Additionally, on the dataset, the YOLOv8 algorithm obtains a mAP50 of 99.5%, improves the precision of 99.9%, and achieves an accuracy of 99.3% for face recognition. Furthermore, the YOLOv8 algorithm offers sufficient speed and accuracy on small faces.