Face recognition for smart attendance system using deep learning

Abstract

Attendance systems using traditional methods take much time and are less efficient. Face recognition is applied by comparing and recognizing the faces in the room with those in the database. However, we can address this issue by using a smart attendance system using face recognition to do attendance automatically and can save time in registering attendance. In this study, we propose a smart attendance system using the face detection model RetinaFace and MTCNN, then the face recognition model using FaceNet and ArcFace. This model will be evaluated using the WiderFace, Essex Faces 94, and Essex Faces95 datasets to evaluate its accuracy and speed. The final model results show that RetinaFace face detection has average precision (AP) results on the WiderFace dataset of 94.20% easy, 93.24% medium, and 83.55% hard, better than MTCNN with AP values of 83.31% easy, 80.32% medium, 83.55% hard. For the combination of face recognition models, FaceNet + RetinaFace obtained as the best combinations model in recognition and speed with a Rank-1 Recognition Rate of 99.114% and recognition speed per image of 118.90 ms.