ArcFace-Based Retinal Biometric Access Control System with Arduino-Servo Door Mechanism and ESP32 IoT Integration

Abstract

This paper presents the design, implementation, and evaluation of an intelligent, embedded biometric access control system that combines deep learning-based facial recognition with physical door actuation. The proposed system leverages the ArcFace recognition model (InsightFace buffalo_l) deployed on a host computer, a USB-connected webcam for image acquisition, an Arduino Uno microcontroller driving a servo motor for door-locking/unlocking, an HC-SR04 ultrasonic proximity sensor for presence detection, Eveready AA batteries for autonomous power, and an ESP32 Wi-Fi/Bluetooth module for future IoT extensibility. A FastAPI backend processes live video, performs motion detection, conducts liveness anti-spoofing, matches face embeddings against a MongoDB Atlas cloud database, enforces role-based time-window access policies, and commands the Arduino via serial communication. A React-based administration dashboard provides real-time monitoring, user enrollment, access logs, and alert management. Experimental evaluation demonstrates a verification accuracy above 96% on a local dataset, sub-second recognition latency, and reliable hardware integration with near-zero false-accept rate under controlled conditions. The system is well-suited for residential, laboratory, or small-enterprise deployment.

Introduction

The text describes a smart biometric door access control system that replaces traditional password, PIN, and RFID-based security methods, which are vulnerable to theft and attacks. It uses face recognition based on deep learning models, specifically ArcFace with InsightFace, to provide accurate and contactless authentication.

The system combines high-accuracy AI with low-cost hardware such as an Arduino Uno, ESP32, SG90 servo motor, and ultrasonic sensor to physically control door locking and unlocking. Recognition and processing are handled by a cloud/host-based FastAPI backend, while the microcontroller only executes hardware actions like opening or denying access.

The backend system performs face detection, embedding generation, motion detection, and liveness checks (to prevent spoofing using methods like blink detection and head movement tracking). It also includes user management, role-based access control, real-time video streaming, and secure data storage using MongoDB.

A React-based dashboard allows administrators to manage users, monitor access logs, and receive real-time alerts. The system also evaluates performance in terms of accuracy, latency, and false acceptance/rejection rates.

Conclusion

This paper presented a complete, end-to-end biometric access control system integrating ArcFace deep face recognition with Arduino-servo door actuation, ultrasonic proximity sensing, and an ESP32 IoT module. The system achieves 96% genuine accept rate with zero photo-replay false accepts across diverse lighting conditions, with a mean end-to-end latency of 348 ms. The FastAPI backend, React dashboard, and MongoDB Atlas backend collectively form a production-grade platform that is readily extensible for larger deployments. Future work will focus on NIR-based recognition, 3D liveness detection, and full ESP32-based MQTT telemetry for remote administration.

References

[1] J. Deng, J. Guo, N. Xue, and S. Zafeiriou, \"ArcFace: Additive Angular Margin Loss for Deep Face Recognition,\" in Proc. CVPR, 2019, pp. 4690–4699. [2] M. Turk and A. Pentland, \"Eigenfaces for Recognition,\" J. Cognitive Neuroscience, vol. 3, no. 1, pp. 71–86, 1991. [3] T. Ahonen, A. Hadid, and M. Pietikäinen, \"Face Description with Local Binary Patterns,\" IEEE TPAMI, vol. 28, no. 12, pp. 2037–2041, 2006. [4] F. Schroff, D. Kalenichenko, and J. Philbin, \"FaceNet: A Unified Embedding for Face Recognition and Clustering,\" in Proc. CVPR, 2015, pp. 815–823. [5] W. Liu, Y. Wen, Z. Yu, M. Li, B. Raj, and L. Song, \"SphereFace: Deep Hypersphere Embedding for Face Recognition,\" in Proc. CVPR, 2017. [6] H. Wang et al., \"CosFace: Large Margin Cosine Loss for Deep Face Recognition,\" in Proc. CVPR, 2018. [7] S. Zhang, X. Zhu, Z. Lei, H. Shi, X. Wang, and S. Z. Li, \"S3FD: Single Shot Scale-Invariant Face Detector,\" in Proc. ICCV, 2017. [8] M. Maheswari and R. Asokan, \"IoT-Based Smart Door Lock System Using Raspberry Pi and Face Recognition,\" Int. J. Adv. Sci. Technol., vol. 29, no. 3, 2020. [9] T. Soukupova and J. Cech, \"Real-Time Eye Blink Detection Using Facial Landmarks,\" in Proc. Computer Vision Winter Workshop, 2016.

Copyright

Copyright © 2026 Selvin David, Hurnish S, Kousalya A. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.