2026-11-03T23:09:58Z https://jmvra.org/oai.php

oai:jmvra.org:JMVRA2610008 2026-10-31 vol4

AI-POWERED REAL-TIME FACE RECOGNITION ATTENDANCE SYSTEM WITH AUTOMATED EXCEL-COMPATIBLE REPORT GENERATION Abhilash Telkar Shivakumar Akshay Narayan Naik Dhanush Kumar M R Gagandeep K L Face recognition ArcFace SCRFD pgvector attendance automation role-based access control. Computer Science and IT Background: Manual classroom attendance is slow, interrupts teaching and is open to proxy marking. Methods: This paper presents an AI-powered, real-time face-recognition attendance system. A browser sends camera frames to a FastAPI backend, which forwards them to an isolated machine-learning microservice. SCRFD detects faces, five-point alignment is applied, and ArcFace (w600k_r50) produces 512-dimensional L2-normalised embeddings. The backend matches these against an enrolled gallery in PostgreSQL with pgvector using cosine similarity, applies a threshold and quality check, and records attendance at most once per student per session. Access is secured with Argon2id, JWT and role-based access control, and a randomised head-movement challenge with proxy alerts adds a lightweight anti-spoofing layer. Results: The system passed 73 automated tests (43 ML, 30 backend), and gallery search takes about 5.9 ms for 1,500 vectors and 19.7 ms for 5,000 vectors. Recognition accuracy (FAR, FRR, Rank-1) is not yet measured because the available dataset has no identity labels. Conclusion: The system is a secure, modular and testable prototype; identity-labelled field calibration is the main next step. JMVRA Publication 2026-10-31 Text Journal Article info:eu-repo/semantics/article info:eu-repo/semantics/publishedVersion application/pdf https://jmvra.org/viewpaperforall.php?paper=JMVRA2610008 https://jmvra.org/papers/JMVRA2610008.pdf Journal of Modern and Versatile Research Approach; ISSN 2984-8911; Vol. 4; Issue 10; pp. a89-a96 en https://creativecommons.org/licenses/by/4.0/ info:eu-repo/semantics/openAccess Copyright © 2026 The Author(s). Published by JMVRA Publication under the Creative Commons Attribution 4.0 International License (CC BY 4.0).