Face recognition access control, from edge device to your own app
SenzrOS runs on RV1126B (Buildroot, up to 150,000 faces) and RK3566 readers. SenzrOne syncs every credential over MQTT on AWS. Your team builds the custom web and mobile apps on top.
What is the Senzr access control SDK? It is a two-part platform. SenzrOS is the edge software for access readers, with a built-in face recognition SDK on RV1126B (supporting up to 150,000 faces on-device) and an Android SDK on RK3566 that works with third-party face recognition engines. SenzrOne is the cloud connector on AWS that synchronizes faces, fingerprints, RFID cards, QR codes, NFC, BLE passes, and UWB codes to every device over MQTT.
Companies get access to SenzrOne on AWS to develop their own custom web and mobile apps, ensuring the user experience carries your brand and facility workflows.
Two Edge Platforms, One Unified Cloud
Select the right hardware foundation for your deployment. Dedicated readers run lean on Buildroot Linux; interactive kiosks run full touch UI on Android.
RV1126B + Buildroot + SenzrOS
A lean embedded Linux build for dedicated access readers, turnstiles, and door controllers with face recognition built directly into the operating system.
- ✓ Built-in face recognition SDK powered by on-chip NPU
- ✓ Up to 150,000 faces stored on the device with sub-35ms matching
- ✓ SenzrOS runtime with native SenzrOne MQTT cloud connectivity
- ✓ Ideal for gates, doors and kiosks that must keep operating at scale
RK3566 + Android SDK
An Android foundation for touch-screen kiosks, visitor registration terminals, and appliances where you want to deploy rich UI apps or a face engine you already license.
- ✓ Android SDK for access control state machines and relay control
- ✓ Bring your own third-party face SDK (ArcSoft, Baidu, Megvii, etc.)
- ✓ Same SenzrOne cloud with unified credential synchronization
- ✓ Suited for interactive kiosks and visitor terminals with rich touch UI
| Platform Feature | RV1126B Platform | RK3566 Platform |
|---|---|---|
| Operating System | Buildroot Linux | Android |
| Edge Software | SenzrOS | Senzr Android SDK |
| Face Recognition Engine | Built-in, up to 150,000 faces (On-chip NPU) | Bring-your-own 3rd-party face SDK |
| Cloud Connector | SenzrOne over MQTT, hosted on AWS | |
| Target Form Factors | Face terminals, boom barriers, speed turnstiles, doors | Self-service kiosks, visitor sign-in, attendance stations |
How a Face Check Works on the Device: 7 Steps
From camera sensor exposure to on-device NPU vector matching, hardware relay actuation, and asynchronous MQTT cloud reporting.
Dual RGB+NIR Capture
Synchronized optical capture across wide-angle RGB and 850nm Near-Infrared sensors for dual-stream illumination invariance.
ISP Glare Suppression
Hardware ISP applies adaptive contrast equalization to neutralize extreme sunlight, nighttime vehicle headlights, and harsh backlighting.
NPU Face Bounding
Neural Processing Unit runs multi-task face detection, generating 5 facial landmark anchor points in under 10 milliseconds.
Infrared Liveness Check
Multi-spectral liveness detection verifies real human skin depth, instantaneously rejecting printed photos, silicone masks, and smartphone screen replays.
512-D Feature Extraction
Quantized deep neural network generates an optimized 512-dimensional INT8 facial embedding vector representing unique biometric identity.
Sub-35ms Local Match
Executes parallel cosine distance search across up to 150,000 local templates on-device. Zero network or internet dependency required.
Relay Trigger & MQTT Event
libgpiod character driver fires the physical gate relay in <1.4ms while dispatching an encrypted JSON event over MQTT to SenzrOne on AWS.
Silicon at a Glance: Edge Processors & Cloud
Senzr supports Rockchip and NVIDIA edge silicon families, paired with high-availability cloud orchestration on AWS.
| Processor Tier | CPU & Architecture | NPU / AI Compute | Operating System | Recommended Role |
|---|---|---|---|---|
| Rockchip RV1126B | Quad Cortex-A53 up to 1.6 GHz | Up to 3.0 TOPS INT8 | Buildroot Linux (SenzrOS) | Dedicated face terminals, turnstiles, 150k offline faces |
| Rockchip RK3566 | Quad Cortex-A55 up to 1.8 GHz | 0.8–1.0 TOPS INT8 | Android | Interactive touch kiosks, visitor registration, 3rd-party face SDKs |
| Rockchip RK3588 | Octa-core (4x A76 + 4x A55) | 6.0 TOPS Tri-Core NPU | Industrial Linux LTS (SenzrOS) | High-throughput vehicle gates, ANPR multi-channel, master edge station |
| NVIDIA Jetson Edge | ARM64 + NVIDIA Ampere/Ada GPU | 20–70+ TOPS TensorRT | Linux Ubuntu / L4T | Multi-camera heavy surveillance, complex edge analytics clusters |
| Cloud Infrastructure | AWS ECS / IoT Core / Aurora | Cloud Orchestration | SenzrOne Cloud Connector | Global credential synchronization, MQTT broker, developer REST APIs |
Storage Planning for 150,000 Faces on RV1126B
Engineered for extreme memory efficiency. Storing 150,000 biometric templates consumes less than 4% of standard flash storage.
| Template Format | Size Per Enrolled Face | Total Storage for 150,000 Faces | Flash Footprint (8GB eMMC) | Offline Search Latency |
|---|---|---|---|---|
| FP32 Unquantized Vectors | 2,048 Bytes (2.0 KB) | ~307.2 MB | ~3.7% of total flash storage | < 50 ms |
| INT8 Quantized Vectors (SenzrOS Default) | 512 Bytes (0.5 KB) | ~76.8 MB | < 0.95% of total flash storage | < 35 ms |
*Note: Estimates represent raw biometric feature templates and vector search indices. Image crops are stored selectively for audit logs with automatic FIFO rotation.
What the Cloud Sends: Credential Sync Specifications
SenzrOne serializes and syncs all 7 credential types over lightweight MQTT payloads with end-to-end TLS 1.3 encryption.
| Credential Type | Synchronized Payload Format | On-Device Verification Target | Local Cache Storage |
|---|---|---|---|
| Face Template | 512-byte INT8 feature vector + liveness threshold metadata | NPU Cosine Distance Engine (>0.82 match threshold) | Encrypted local biometric database |
| Fingerprint | ISO/IEC 19794-2 minutiae template (256–512 bytes) | Capacitive / Optical fingerprint sensor match | Local SQLite Whitelist |
| RFID Card | CSN / UID hex string + facility code (Mifare / DESFire / HID) | Wiegand 26/34 / RS485 RFID reader controller | Instant Indexed Hash Map (<0.5ms) |
| Dynamic QR Code | Time-based HMAC-SHA256 token string with 60-second validity | Onboard camera QR decoder + shared secret key check | Offline HMAC Validator |
| NFC Credential | ISO 14443 Type A/B card UID or encrypted virtual token | NFC reader antenna controller | Local Token Whitelist |
| BLE Mobile Pass | Rotating AES-128 proximity beacon payload | Bluetooth 5.0 Low Energy transceiver daemon | Dynamic Session Cache |
| UWB Code | Cryptographic distance threshold token + session seed | Ultra-Wideband ranging transceiver (<10cm precision) | Ranging Security Supervisor |
Reader Hardware Interfaces to Design Around
Comprehensive hardware pinouts and industrial peripherals supported natively by the SenzrOS hardware abstraction layer.
| Interface Category | Physical Connectors | Linux / Android Driver | Access Control Peripheral |
|---|---|---|---|
| Cameras | Dual MIPI-CSI (2-lane / 4-lane) or USB 2.0/3.0 UVC | V4L2 Video Subsystem + Rockchip ISP Driver | Synchronized RGB + 850nm NIR biometric sensor pair |
| Display & Touch | MIPI-DSI, LVDS, or HDMI (Up to 1080p / 4K) | DRM / KMS + Android SurfaceFlinger | User visual feedback, greeting, QR display, touch keypad |
| Serial Bus | RS485 (Galvanically Isolated) & RS232 | Linux TTY Serial Driver (/dev/ttyS*) | External card readers, biometric slaves, elevator controllers |
| Industrial CAN | CAN 2.0B / CAN-FD Terminal Blocks | SocketCAN Kernel Network Stack | High-reliability barrier motors, automated vehicle gates |
| Relay Outputs | Form-C Dry Contacts (NO/NC/COM) | libgpiod Character Device Driver | Speed turnstiles, maglocks, electric strikes, boom barriers |
| Safety Inputs | Optocoupled Digital Inputs (DI) | Kernel GPIO Edge Interrupts (<1.4ms response) | Optical safety infrared beams, fire alarm release triggers |
| Legacy Wiegand | Data0 / Data1 differential pairs | Kernel GPIO Bit-Bang / Dedicated Wiegand Controller | Integration into existing 3rd-party building panels |
ANPR-Based Access Control Under One Policy
Manage vehicle license plates and human biometric credentials in the same unified cloud policy on SenzrOne.
Dual-Factor Vehicle & Driver Clearance
For high-security industrial depots, residential townships, and commercial freight hubs, Senzr allows boom barriers to open only when both vehicle and driver credentials match the authorized roster:
- ✓ Real-time ANPR: Sub-15ms recognition on high-speed camera streams
- ✓ Co-Located Driver Biometrics: Driver face verified simultaneously via roadside terminal
- ✓ Single Security Ledger: One audit record logged to SenzrOne on AWS containing vehicle plate, driver identity, timestamp, and gate telemetry
Build Your Own Apps on SenzrOne
Your developers work against modern cloud APIs and MQTT topics, not raw hardware drivers. Build custom tenant apps, admin portals, and visitor management systems.
Speed Race: SenzrOS vs Traditional Cloud
Experience firsthand why cloud-dependent access control fails during peak rush hours or network cuts.
Live Access Barrier Benchmark Simulator
Benchmarking on-device face recognition against cloud API queues.
Hardware Command Console: senzr-ctl
Experience the onboard diagnostic utility used by systems engineers for real-time telemetry, relay testing, and offline rule verification.
senzros-node-01:~$ senzr-ctl status
[OK] Hardware Model : Rockchip RV1126B / RK3588 Edge Station
[OK] Face Vector DB : 150,000 Identities Loaded · Index: INT8 HNSW
[OK] NPU Cores : 3.0 TOPS Active · Temp 37.0°C
[OK] Barrier Relay : Pin GPIO1_D4 · State: DISARMED
[OK] Safety Beam : Pin GPIO3_A2 · State: CLEAR (Interlock Armed)
[OK] Cloud Uplink : MQTTS tls://iot.senzr.in:8883 (SenzrOne on AWS Connected)
senzros-node-01:~$ _
Frequently Asked Questions
Key technical and architectural answers for device OEMs, system integrators, and enterprise software teams.
Plan Your Access Control Rollout
Tell us your sites, readers and credential types. Our engineering team will recommend the platform (RV1126B Buildroot or RK3566 Android) and the integration path on AWS.