Comparative Analysis of FaceNet and ArcFace in Minimizing False Positives for Enhanced Access Control Security

Face recognition technology is crucial for enhancing security in access control systems, yet false positives remain a significant challenge. This paper compares two leading pre-trained face recognition models, FaceNet and ArcFace, to determine which model better minimizes false positives in access control scenarios. The study involves testing the models under various real-world conditions, including different distances, lighting levels, and the presence of multiple subjects, to simulate actual operational environments. Experiments using live stream video from a standard laptop webcam with subjects not included in the training dataset assessed each model's ability to accurately recognize unfamiliar faces. The results indicate that FaceNet outperforms ArcFace in real-world conditions, exhibiting lower false positive rates across varying distances and lighting conditions, while ArcFace shows consistent performance in multiple person detection, suggesting reliability in controlled settings.