Answers are grounded in this paper and cite their sources.
Face recognition has emerged as an effective biometric authentication technology for intelligent access control systems because of its contactless operation, convenience, and security advantages. Although deep learning has substantially improved recognition performance, the suitability of different models for real-time deployment on embedded platforms remains insufficiently explored. This study presents a real-time, face recognition-based smart door lock system implemented on a Raspberry Pi 5 and compares three face recognition models: FaceNet, FaceNet512, and InsightFace. The system integrates a webcam, a relay module, and an electromagnetic door lock to provide automated access control for authorized users. Facial images were collected from 20 individuals, yielding 400 training images and 200 authorized-user testing images. Model performance was evaluated using accuracy, precision, sensitivity, specificity, F1-score, latency, frames per second (FPS), and recognition robustness at different distances. InsightFace achieved the highest performance, with 100% accuracy, precision, sensitivity, specificity, and F1-score on the testing dataset. During real-time testing, InsightFace maintained authorized-user recognition rates above 97% at all tested distances and achieved a 100% unauthorized-user rejection rate. It also recorded the lowest average latency (0.243 s) and the highest average processing speed (4.153 FPS), demonstrating its suitability for embedded smart door lock applications.
No takes yet. Share an insight, caveat, or question.
Halid et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: