Skip to content
Vision AI
Technology

Engineered for the physical world.

Real cameras, real lighting, real latency constraints. Our architecture starts from the conditions of deployment rather than the benchmark.

event · vision.intrusion● live
{
  "type": "intrusion",
  "camera": "cam-04",
  "zone": "A-restricted",
  "class": "person",
  "confidence": 0.97,
  "track_id": 7183,
  "latency_ms": 214,
  "evidence": "clip://…/7183.mp4",
  "actions": ["notify:ops", "vms:bookmark"]
}
How it works

From pixels to decisions in under a second.

A streaming architecture that captures, infers at the edge, correlates across cameras, and acts through the systems you already run.

  1. 01

    Capture

    Ingest from any camera, VMS, or sensor over RTSP/ONVIF, with health monitoring and calibration.

    • RTSP / ONVIF / VMS
    • Decode & normalise
    • Calibrate & map zones
  2. 02

    Edge Inference

    Run optimised models on-site to detect, track, and recognise with millisecond latency.

    • Detection & tracking
    • Recognition & OCR
    • Anonymise at source
  3. 03

    Analytics

    Correlate events across cameras and time into metrics, anomalies, and context-rich alerts.

    • Zones, counts, dwell
    • Anomaly detection
    • Cross-camera correlation
  4. 04

    Action

    Route to people and systems: alerts, workflows, device control, and enterprise integrations.

    • Rules & escalation
    • Device & gate control
    • APIs & connectors

Deploy anywhere

Edge, on-prem, cloud, or hybrid.

  • Edge
    Appliances or your hardware on-site; raw video never leaves the premises.
  • On-premises
    Full platform inside your data centre, including air-gapped environments.
  • Cloud
    Managed platform for aggregation, dashboards, and model operations.
  • Hybrid
    Edge inference with cloud governance and cross-site analytics.
Model stack

Purpose-built models for the physical world

A layered model stack covers the full perception task, from locating objects to understanding behaviour, tuned for the camera angles, lighting, and clutter of real deployments.

  • Object detection

    People, vehicles, PPE, packages, hazards, and custom classes with configurable confidence.

  • Multi-object tracking

    Persistent identities across frames and occlusions for counting, dwell, and trajectories.

  • Re-identification

    Anonymous appearance embeddings to follow movement across cameras without biometrics.

  • Recognition & OCR

    Plates, codes, and opt-in face recognition with liveness and audit controls.

  • Pose & behaviour

    Posture, falls, and activity recognition for safety and operational use cases.

  • Scene understanding

    Fire, smoke, spills, crowd density, and environmental conditions.

Edge compute

Intelligence where latency and bandwidth matter

Edge-first architecture keeps raw video on-site, reduces cost, and guarantees response times. Cloud or data-centre resources aggregate, govern, and report.

  • Accelerator-aware runtime

    Automatic use of GPU, NPU, and VPU hardware with precision tuning.

  • Model compilation

    Quantisation and compilation pipelines for each target without retraining.

  • Offline resilience

    Store-and-forward and local rules keep sites protected through outages.

  • Fleet operations

    Zero-touch enrolment, staged rollouts, health telemetry, and rollback.

Real-time pipeline

Built for continuous, high-throughput streams

A streaming architecture decodes, infers, correlates, and dispatches events in parallel, so adding cameras scales horizontally rather than slowing everything down.

  • Hardware decode

    H.264 / H.265 decoding on silicon to free compute for inference.

  • Event correlation

    Fuse detections across cameras and time into single, de-duplicated events.

  • Back-pressure & QoS

    Graceful degradation prioritises critical analytics under load.

  • Observability

    Per-stream metrics, model health, and end-to-end latency tracing.

Integration

Open by design

3Cubicles fits the systems you already run: cameras, VMS, access control, building management, ticketing, BI, and messaging, through standards and well-documented APIs.

  • Camera & VMS standards

    RTSP, ONVIF Profile S/T, and native VMS connectors.

  • Events & messaging

    Webhooks, MQTT, and streaming events for real-time consumers.

  • APIs & SDKs

    REST APIs plus Python and JavaScript SDKs with typed schemas.

  • Enterprise systems

    SSO, ITSM, BI, data warehouses, and industrial protocols such as OPC-UA.

Privacy & security

Trust engineered into every layer

Vision AI only earns adoption when it respects people. Privacy and security controls are built into the platform rather than bolted on.

  • Anonymisation at source

    Edge masking and metadata-only modes so identities never leave the site.

  • Retention & residency

    Configurable retention, regional storage, and legal-hold controls.

  • Access & audit

    SSO, RBAC, and immutable audit logs for every view, export, and change.

  • Secure supply chain

    Signed models and updates, encrypted storage, and hardware identity.

Standards & protocols

Speaks the languages your infrastructure already uses.

RTSPONVIF Profile S/TH.264 / H.265RESTWebhooksMQTTOPC-UASAML / OIDCSCIMONNXTensorRTKubernetesOCI containersS3-compatible storageSyslog / SIEM

Support for specific vendors and versions: ONVIF Profile S/T cameras from major vendors (Axis, Hikvision, Dahua, Hanwha, Bosch); Milestone XProtect and Genetec Security Center via plug-in; RTSP with H.264 / H.265.

Perceive. Understand. Act.

Talk to our engineers.

Architecture reviews, proof-of-concept scoping, and integration planning with the people who build the platform.

Request a technical sessionContact Sales