Video Surveillance Analytics

Defence · 2016-2019

PublicRepresentative · synthetic data
  1. 1

    Run a scenario on the Wall

    The Wall is the product surface - a multi-camera grid with restricted-zone overlays and per-camera status. Pick a scenario (a vehicle at the gate, a pedestrian over the fence, a shadow or animal false positive, a multi-object cluster, or a partial occlusion) and Step the clock to advance the simulated camera frame by frame.

    Watch an intrusion build: the dwell counter (e.g. 2/3) has to fill before an alert fires - a single odd frame never raises one. Equally instructive are the scenarios where nothing fires: a shadow dropped by the confidence floor, an animal ignored by a zone's class allow-list.

  2. 2

    Handle alerts

    When an alert fires it lands in the priority feed with a snapshot and its camera, zone, class and time. Acknowledge it, dispatch a patrol, resolve it, dismiss it, or flag it as a false alarm. A false alarm feeds the False-alarm review in the console. An unhandled alert escalates up the chain on timeout, and later frames can confirm a resolution.

    Open "How this was decided" on any alert to see the full ten-stage trace in the inspector - which single stage spent, what the change gate skipped, and how the dwell window confirmed the intrusion.

  3. 3

    Tune it in the console

    The Console is role-gated. The owner-only Configure section is a two-layer baseline ⊕ overlay: adjust the detection floor, the per-zone class allow-lists, the dwell window, the alarm severities and the operator routing - every edit drives the engine live and is audited, with a one-click reset to baseline. The inspector carries an on-prem / data-sovereignty note, and everything is labelled simulated and synthetic. The project-page walkthrough video narrates the full tour.