Real-Time Multi-Face Attendance Tracking System Using Deep Learning

Abstract

The conventional attendance management systems are prone to errors, time-consuming, and prone to proxy fraud, which are the basic weaknesses restricting their efficiency in contemporary educational institutions. The following paper discusses the Smart Multi-Face Attendance System, a web-based, 100% automated, full-fledged deep learning system capable of tracking student attendance in real time without physical interaction and with additional equipment. It uses the InsightFace buffalo_l model ( ArcFace R100 + RetinaFace ) to produce 512-dimensional face embeddings, and is matched by identity using a cosine similarity between simultaneous live video streams. The dual-camera system used keeps track of entry and exit with a way of loading accurate time-present computation per student. Findings indicate 99.83% accuracy on LFW benchmark, 25-30 FPS graphics card performance and end to end attendance computation time of less than 60 seconds. The system surpasses all major features of the current paper based, RFID and fingerprint systems and does so with no extra hardware needed.