Design Approach of Electronic Attendance-Absence Recording System using Multi-User Face Recognition

In this paper, we propose the design and service scenario of an electronic attendance-absence recording system (EAARS) based on face recognition that is convenient to use and cannot be fraudulently attended. We suggest the design of the system for commercialization that utilizes state-of-the-art (SOTA) deep learning facial recognition technology, which offers high accuracy and fast speed. The system is designed to simultaneously recognize multiple users' faces and store the attendance and absence of each person. The electronic attendance-absence recording system consists of a server, a mobile platform for users or students, a personal computer (PC) platform for lecturers or teachers, and a multi-user face recognition module. It is configured in the form of TCP communication using the JSON file format with the web server. To implement the multi-user face recognition module, we utilize the SOTA technologies in face detection and face recognition, namely RetinaFace and ArcFace. Various backbone networks such as ResNet50, MobileNet V2, MobileNetV3, and MobileViT are used for training and we compare the recognition results and speed to select the appropriate model. The WiderFace database is used for developing face detection module, while MS-Celeb-1M and LFW are used for face recognition. In addition, we use TensorRT to optimize the trained model to improve the speed of the Multi-User Face Recognition module. We believe that it is necessary to specify the service scenario in detail for how to deal with the claims of users or students in case of misidentification in the system in the future.