CPU-Optimized Real-Time Face Recognition For Automated Attendance Management: A Mediapipe-Based Approach With Transparent Audit Logging

Automated attendance management in educational and institutional settings demands solutions that are simultaneously accurate, cost-effective, and deployable without reliance on external infrastructure. This paper presents a face recognition-based attendance system built upon Google's MediaPipe Face Mesh framework, designed specifically for operation on standard central processing units (CPUs) without requiring dedicated graphics processing units (GPUs) or cloud connectivity. The proposed architecture employs a four-layer pipeline: real-time video capture, 468-landmark facial feature extraction, 128-dimensional embedding generation with L2 normalization, and cosine similarity matching against a pre-enrolled reference database. Attendance records are persisted in structured comma-separated values (CSV) format with built-in duplicate prevention and timestamp logging. Experimental evaluation across a cohort of N = 70 participants in controlled, variable-illumination, and real-world classroom environments demonstrates recognition accuracy of 97.4%, a False Acceptance Rate (FAR) of 0.8%, and a False Rejection Rate (FRR) of 2.1%, with median per-frame inference latency of 38 ms on commodity hardware. The system operates fully offline, requires no proprietary hardware, and produces human-readable audit trails, making it suitable for institutions with limited IT infrastructure.