Edge side
- Camera capture and local feature extraction
- Offline-safe spool queue with retry behavior
- Signed payloads and device authentication
- Hardware options spanning Raspberry Pi, Jetson, and standard camera inputs
About FaceTrack
The product exists to make face recognition deployable, explainable, and commercially usable.
The about page now keeps one clean narrative: local capture, secure sync, cloud orchestration, and operator trust.
Cloud side
- Tenant-scoped onboarding and camera inventory
- Queue-based matching with confidence bands
- Unknown person alerts, match history, and audit trails
- Privacy and retention guardrails for biometric operations
Why this structure
Good real-world accuracy depends on the whole deployment, not only the model.
That is consistent with current market messaging from Face-Six and NCheck: camera quality, enrollment quality, thresholds, and workflow design matter as much as the recognition engine.
Camera quality
Lighting, angle, motion blur, and pixel density will wreck matching before any dashboard redesign can save it.
Enrollment quality
Clean samples and multi-pose coverage reduce bad matches and manual review churn.
Operational rules
Confidence thresholds, alert windows, and tenant policies decide whether alerts are useful or just noise.