Real-Time Face Recognition for Automated Attendance Management: A Deep Learning Approach Using FaceNet and ArcFace

Abstract

Manual and biometric attendance systems used in educational institutions suffer from proxy attendance, hygiene concerns, and operational inefficiency. This paper presents a real-time, contactless attendance management system built on deep learning-based face recognition. The proposed pipeline integrates Multi-task Cascaded Convolutional Networks (MTCNN) for face detection with a FaceNet embedding network fine-tuned using ArcFace loss on a custom institutional dataset of 120 students. The system achieves 98.7% recognition accuracy on the test set, processes up to three simultaneous faces in under 200 ms, and marks attendance with an end-to-end latency below 500 ms per student. Experiments are conducted on a custom 4,200-image dataset collected across three classroom environments with varied lighting conditions. The system operates entirely on-premises, ensuring compliance with India's Digital Personal Data Protection (DPDP) Act. Results demonstrate that the proposed approach outperforms comparable open-source solutions and is deployable on commodity GPU hardware at costs accessible to resource-constrained institutions.

Keywords