Face Recognition for Student Attendance Using Mtcnn and Pre-Trained Facenet

Abstract

Attendance management remains a critical component in both industrial and educational settings. Traditional attendance methods based on manual signatures are often inefficient, prone to errors, and vulnerable to impersonation. In contrast, biometric systems, particularly those based on face recognition, offer a promising alternative. This paper presents a novel attendance system using deep learning techniques that combine a Multitask Convolutional Neural Network (MTCNN) for robust face detection and alignment with the FaceNet model for generating discriminative 128-dimensional facial embeddings. The system automatically captures, verifies, and logs attendance data into an Excel sheet in real time. On a small dataset obtained from real-world work and higher education environments, our system achieved an overall accuracy of 95% under normal conditions. Additionally, we discuss system performance under challenging scenarios such as overlapping faces and intense lighting, and we propose future enhancements for dynamic retraining and improved classification strategies.