Detection & continuous tracking
Locks onto faces stably, providing reliable input for quality gating, feature extraction and identity verification.
RK3576 EDGE AI FACE ENGINE
Built and tuned around the FA8 & FA10 terminals and the 6-TOPS RK3576 NPU. Detection, anti-spoofing and identity decisions all run locally — no cloud round-trip, fully offline.
01ENGINEERED ON DEVICE
Edge-first R&D and device tuning
Our algorithms are developed and co-tuned directly on FA8, FA10 and the RK3576 platform. Through 8 generations of engineering iteration on tens of millions of compliant multi-scenario samples, recognition, quality gating and binocular liveness operate in one unified edge engine.
Explore core capabilitiesCore model inference runs on the dedicated NPU
The compute foundation for edge vision workloads
Continuous algorithm and model engineering
With or without a network, on-site recognition keeps running
02ONE PIPELINE · MULTIPLE SIGNALS
Recognition and anti-spoofing in one edge pipeline
Face detection, quality & occlusion gating, identity search and binocular liveness run in concert — key model inference handled by the on-device NPU.
Explore algorithms & modelsBoth spectra processed together — the pipeline runs entirely on the device
Locks onto faces stably, providing reliable input for quality gating, feature extraction and identity verification.
Fuses pose, sharpness, lighting and facial visibility to keep low-quality frames out of the decision.
Fuses RGB with 850 nm near-infrared to stop photo prints and screen-replay attacks.
Edge-side search optimized for large galleries, so verification stays fast at scale.
03MEASURED IN TESTS
Representative results from R&D evaluation
Metrics come from internal R&D evaluation. Hardware, lighting, distance and footfall all affect real-world performance — validate on target devices in the field.
View benchmarks & complianceEdge-side person search for large campuses and organizations.
Recorded under the corresponding R&D evaluation conditions.
Consistency of results across demographic groups in testing.
Rank correlation against an international reference quality yardstick.
Note: figures above are drawn from SUNTEK internal R&D evaluations under fixed test thresholds and do not constitute a performance guarantee for every device, capture condition or deployment. Validate with your own hardware and field tests before production.
04BUILT AROUND REAL HARDWARE
Tuned for real terminals
Developed and joint-debugged around the terminals’ dual cameras, illuminators, binocular imaging and NPU inference, so the algorithm fits real operating conditions.
Explore devices & solutions
The all-round dual-spectrum edge terminal for everyday attendance and high-throughput on-site identity checks.

Jointly calibrated around a precision binocular baseline, adaptive IR illumination and strict anti-spoofing.
Hardware-anchored engineering. Published metrics describe our R&D direction; final results in your project still depend on mounting height, lighting, lens and footfall — tune on site.
05DESIGNED FOR REAL-WORLD FLOW
For real operational sites
From campus attendance to production lines, fixed terminals handle identity checks — and keep running locally when the network doesn’t.
Employee verification, offline clock-in and intranet workflows.
02 / CLEANROOM & CAMPUSReliable masked and bare-face access on fixed terminals.
03 / INDUSTRY & FINANCEAir-gapped edge recognition and deployment in restricted networks.
06FULL-STACK PRODUCT OVERVIEW
Deep-dive into SUNTEK Face AI
Explore how SUNTEK Biometric Engine unites multi-model edge NPU intelligence, FA8 & FA10 binocular terminals, independently verified benchmarks, and our v2.6.5 cloud-edge SDK & self-service licensing portal.
Unified RK3576 NPU architecture combining face detection, 98-pt landmark alignment, wearing & ISO 29794-5 quality gating, RGB + 850nm NIR liveness, and adaptive subspace matching.
Explore algorithms & modelsCo-engineered RK3576 binocular terminals with 9:16 touch displays, zero-lag dual-spectrum cameras, adaptive 850nm IR fill light, and solutions for campuses, cleanrooms, and banks.
Explore devices & solutions99.91% bare / 99.97% masked 50K open-set retrieval, 0.00% dual-spectrum PAD spoof pass rate, 0.25% demographic parity spread, and native EU AI Act, GDPR & ISO alignment.
View benchmarks & complianceUnified edge AAR, multi-arch Node.js server SDK, and 2.5 MB mobile capture across Android, iOS, HarmonyOS NEXT, Web WASM, Flutter, RN & UniApp with <1ms 50K search and QR/offline activation.
Read SDK & licensing guideSUNTEK FACE AI