Facial Biometrics Access Control
A secure facial recognition access control system deployed across multiple corporate office locations, replacing legacy RFID cards with biometric verification and providing real-time access logs and anomaly alerts.
Results
The Challenge
The client, a corporate facilities management group, was responsible for physical security across six office sites. RFID access cards were being shared, lost, and cloned at an increasing rate. A security review identified that 23% of access events could not be attributed to the registered cardholder. They needed a system that tied access to the person, not a card — without creating a friction-heavy experience for the 400+ daily users.
Our Solution
We designed a distributed facial recognition system using purpose-built hardware nodes at each entry point. Each node ran a local inference engine for sub-second recognition without sending raw imagery to a central server. A central management platform provided enrolment, access policy management, and real-time audit logs. The system operated in hybrid mode — facial recognition as the primary factor, with PIN as fallback for edge cases such as mask wearing.
Privacy by Design
Facial recognition in the workplace raises legitimate privacy concerns. We engaged a data protection solicitor before the technical design began. The system stored facial embeddings — mathematical representations of facial geometry — not photographs. Embeddings were encrypted and stored only on-premises. No data was shared with third parties. Staff were informed, consented, and given a right to opt out (with alternative access provision). An ICO registration and DPIA were completed before go-live.
Edge Inference Architecture
Sending camera footage to a central server for processing creates latency, bandwidth costs, and a single point of failure. We deployed the recognition model directly to Raspberry Pi 4 nodes at each door, containerised with Docker. Each node processed faces locally and sent only a pass/fail signal with an anonymised user ID to the central system. This reduced average entry time to under one second and meant the system continued to function during network outages.
Enrolment and Management
The management platform allowed HR administrators to enrol new staff using a standard webcam — a 30-second process that captured multiple angles. Access policies assigned different entry permissions by time of day, day of week, and site. Temporary access for contractors could be granted with an automatic expiry date. The audit log provided a full, timestamped record of every access event, exportable for compliance reporting.
Fallback and Edge Cases
No biometric system is infallible. We designed the fallback carefully. If recognition confidence fell below 0.92, the system prompted for a six-digit PIN. Mask detection was integrated — a masked face triggered PIN mode automatically rather than failing. The security desk received an instant alert for any failed authentication, with a camera snapshot for manual review. False acceptance rate in production testing was below 0.01%.
Outcome
Within three months of full deployment, unauthorised access events had fallen by 94%. The security review that had previously identified 23% unverifiable access events recorded 1.4% in the subsequent audit — the residual figure attributable to contractor visits. Average entry time was 0.8 seconds per person. Staff feedback was positive, with the majority preferring the hands-free experience to searching for an RFID card.
Have a similar project in mind?
Tell us what you are building. We will give you an honest assessment and a free quote.
Start a Conversation