An Efficient Raspberry Pi-Based Face Recognition System for Smart and Privacy-Aware Student Attendance

Abstract

Traditional attendance methods are slow and fraud-prone. This project implements a face recognition-based attendance system on Raspberry Pi 4, automatically detecting and recog nizing students at classroom entry and logging attendance in real time via a web interface. It uses MediaPipe BlazeFace, MobileFaceNet, ONNX Runtime, Flask, and SQLite, operating fully on-device without cloud dependency. The pipeline includes detection, tracking, align ment, quality filtering, embedding extraction, cosine similarity matching, and multi-frame voting. Evaluation achieved 24.3 ms latency, 95.8 percent accuracy, 0.1 percent false ac ceptance rate, and 110–140 MB memory usage, confirming viability for embedded real-time deployment.