A deep learning based approach for student attendance recognition

Abstract

Traditional attendance recognition methods - such as using cards or signing into lists are susceptible to fraud - often consumes considerable time and effort in managing and processing information. With advancing technology and increasing security requirements, modern attendance recognition methods offers higher accuracy and convenience for organizations. This paper proposes an ap-proach utilizing DeepFace technology and the deep learning architectures (Mo-bileNet, ResNet) to face detection, face recognition and face anti-spoofing. The proposed approach is experimented on the existing dataset within the research community (CelebA-spoof) and on a dataset (named CUSC-Student) built by ourselves; and is evaluated by accuracy and F1-Score metrics. From the experi-mental results, this approach also identifies the suitable model that achieves the high accuracy while requiring less the computing resource. Therefore, it can be applied in practice, specifically for recognizing student attendance at our divi-sion.