DESIGN AND IMPLEMENTATION OF GENERATIVE ADVERSARIAL NETWORK AND YOLOV5 ALGORITMA FOR MASKED FACE RECOGNITION IN PRESENCE MACHINE

• Main Author: Aliya Roostiani, Hasna
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/80885
• Institution: Institut Teknologi Bandung

The COVID-19 pandemic has driven the development of touchless and automatic face recognition technologies. In an effort to control the spread of the virus, the study of masked face detection and recognition is a relevant solution. To prevent the transmission of the COVID-19 virus, a system without human contact is needed, capable of analyzing and identifying individuals wearing masks. This research focuses on face verification that determines whether the faces belong to the same person and classifies the faces into specific identities for the attendance process. This research is built and designed on the case of imperfectly exposed faces, evaluating the capabilities of Generative Adversarial Networks (GANs) and the YOLOv5 algorithm by looking at the results of its standard metrics, including loss, accuracy, precision, recall, and f1-score. The performance achieved in the masked face recognition system for real-time implementation is accuracy = 94.61%, precision = 94.33%, recall = 97.01%, and f1-score = 93.05%. On the other hand, using the validation set by GAN resulted in accuracy = 95.71%, precision = 95.87%, recall = 96.65%, and f1-score = 95.15%. The training loss achieved in this research is 20.6 and the validation loss achieved is 33.4. The YOLOv5 algorithm is implemented in the detection process to obtain the bounding box of the registered face. Meanwhile, the GAN training method is used for masked face recognition which has been integrated into the designed attendance machine. The proposed system successfully detects names with a high level of accuracy despite the use of masks.