NovaEyeD

Production-ready face recognition for embedded products

NovaEyeD delivers production-ready face recognition for edge devices with fully on-device detection, liveness, enrollment, and matching — no cloud or connectivity required.
Demonstrated and deployed across the embedded silicon ecosystem.
Accuracy
0 %
Industry baseline for on-device recognition is typically 93–95%. NovaEyeD exceeds this without a cloud round-trip.
inference
1000 ms

Fast enough for real-time, continuous recognition at the entry point — no queuing, no delay.

SRAM
< 20 MB
Fits within MCU-class memory budgets — no application processor required.
network round-trips
0
Fully offline. No dependency on connectivity, and no biometric data ever leaves the device.

Why teams license NovaEyeD instead of building

Building production-grade face recognition from scratch including detection, liveness, alignment, feature extraction, 1:N matching, and drift handling typically takes embedded teams 18+ months to achieve commercial-grade accuracy and robustness. NovaEyeD delivers the entire pipeline, optimized for your chosen silicon platform.

Featured in STMicroelectronics' Edge AI Suite case study library.

NovaEyeD is featured as a production reference for facial recognition on STM32N6 silicon. Joint demonstrations and benchmarking are publicly available through ST.

NovaEyeD on STM32N6: Edge Face Recognition at ~40ms.

A recorded technical session walking through the NovaEyeD pipeline, performance characteristics, and deployment integration. Hosted with STMicroelectronics.

Demonstrated across the silicon ecosystem.

Live-demonstrated at Embedded World on hardware from STMicroelectronics, Infineon, Synaptics, and Alif Semiconductor – running production workloads on each platform.

Built for embedded products that need face recognition

NovaEyeD is licensed depending on what you’re shipping. Two paths, both with engineering support included.

Skip the build. Ship the product.

If you’re shipping a product with facial recognition – smart locks, access control, kiosks, appliances, industrial equipment – NovaEyeD removes the hardest 6-12 months of your roadmap.

We deliver a production-validated model running on your chosen silicon, integrated into your firmware stack, with support and maintenance to keep it running across product variants and silicon revisions, all while mitigating model drift.

Drive design wins. Win attach.

NovaEyeD gives your field teams a working facial recognition reference customers can demo, deploy, and ship – without an internal AI team or a months-long porting project.
We tune NovaEyeD to your silicon, package it with your SDKs, and stand behind it with engineering support your sales team can offer to design-win customers.

A full face recognition pipeline. Not just a model.

Most “edge face recognition” offerings are a single model – a detector, or a matcher, or a feature extractor. Productionizing them means stitching together the missing stages yourself, optimizing each one for your silicon, and validating the seams. 

NovaEyeD ships the full pipeline. Registration, liveness, recognition, and runtime maintenance – all designed to run together, on-device, within MCU-class memory budgets. 

Pipeline Capabilities

3 Recognition Modes

Closest-face priority for single-user devices, zone-based detection for defined areas, all-faces-in-frame for multi-person scenarios. Configurable per deployment.

Passive Liveness Detection

Camera-based anti-spoofing rejects photo, video, and mask attacks. Runs entirely on-device – no additional sensor required. For deployments needing higher assurance, NovaEyeD also supports time-of-flight sensor input as a secondary signal.

Multi-Face Storage with 1:N Matching

Many registered faces stored on-device, with scalable 1-to-many matching. Storage capacity scales with available flash.

Periodic Embedding Update

User profiles refresh over time as appearance naturally changes – glasses, facial hair, age – without requiring re-enrollment.

Real-World Robustness

Validated across glasses, hats, masks, accessories, lighting extremes (low light through direct spotlight), head pose up to ±30º, and effective detection range from 0.5m to 3m.

Registration

One-time per user. Enroll on-device, via a mobile companion app, or server-side. Embeddings are securely transferred and stored locally – never as raw images.

Detect Face
Capture Angles
Quality Check
Extract Embeddings
Store Profile

Recognition

Runtime, per frame. ~40ms core inference end-to-end. Inference Breakdown: ~10ms detection + ~30ms recognition

PIPELINE

What a NovaEyeD license includes

Most face recognition vendors sell you a model. We sell you the model, the engineering, and the relationship.

Built for the regulatory frameworks face recognition lives under

NovaEyeD’s on-device architecture means biometric data never leaves the device — closing the compliance gap that disqualifies cloud-based face recognition from regulated deployments.

On-Device Processing

100% local inference. Zero network round-trips. Works fully offline.

EU AI Act – Biometric Provisions

Designed for compliance with Annex III high-risk requirements when deployed within controlled use cases.

GDPR-Aligned (EU)

No transfer of biometric data off the device. Data minimization by architecture.

BIPA / CUBI / MyHealth Aligned

State biometric privacy laws addressed by on-device design — no third-party processor in the data flow.

Local Biometric Storage

Face embeddings stored on-device only. Raw images never persisted.

No Cloud
Dependency

Air-gapped deployment supported. No vendor cloud, no third-party API, no recurring telemetry.

Compliance posture depends on your deployment context, your data handling policies, and applicable law. We support your compliance program — we don’t replace it. Talk to our team for deployment-specific guidance.

Built for the products embedded teams actually ship

NovaEyeD deploys across consumer, industrial, automotive, and access control applications — any product where on-device face recognition replaces a cloud round-trip, a keypad, or a key.

Smart Home Entry

Smart Home Entry

Residential smart locks and doorbells. Faces unlock entry instantly for known residents. All biometric data stays in the home.

Industrial Operator Sign-In

Industrial Operator Sign-In

Authenticate operators at the machine. Restrict equipment to trained personnel. Reliable under gloves, hard hats, and shift-change pace.

Access Control

Access Control

Office entry, secure rooms, vaults, lockers. On-device matching means no network dependency and no cloud-side identity database to breach.

Kiosks & Self-Service

Kiosks & Self-Service

Retail checkout, hospitality check-in, healthcare patient kiosks. User-aware personalization without storing biometric data off-device.

Vehicle & Driver Systems

Vehicle & Driver Systems

Driver identification, personalized vehicle settings, fleet operator authentication. Local inference fits automotive privacy and latency requirements.

Appliance Personalization

Appliance Personalization

User-aware smart appliances — coffee machines, refrigerators, smart TVs. Face presence triggers user profiles without an app or account.

Use of NovaEyeD must comply with applicable law and your internal privacy policies. Talk to our team for deployment-specific guidance.

Ready to license NovaEyeD?

Initial scoping is free. NDA available before any technical discussion. Most engagements move from first conversation to signed agreement in 2–4 weeks.
Scroll to Top