Detection
How camera events are reviewed in Virtik: the camera wall, the review states, how an operator or a rule turns a detection into an incident, and what has not been verified.
- Version
- 0.2
- Dernière mise à jour
- Disponibilité
- Partiellement disponible
Parcourir la documentation
Sur cette page
Ce document n'est actuellement publié qu'en anglais. Lire la version anglaise de référence
Detection is where camera events are seen, checked and acted on. Events arrive from an analysis pipeline, an operator reviews each one, and a reviewed event can become an incident. Detection reads the same site model as the rest of Virtik: on the Live wall, a camera that has been placed in a Virtual Site shows the zone, level and building it covers.
The four tabs
| Tab | What it is for |
|---|---|
| Live | A wall of camera tiles, with health, analysis status and per-camera controls. You choose which cameras to put on the wall, from any site, and the wall shows only as many as its layout holds. Cameras are added here. |
| Map | Sites, cameras, events and camera health on a map, with a Fit all control. |
| Events | The list of detections, with filters, a review queue, a CSV export and a detail panel where a verdict is recorded. A Live / Not live indicator shows whether the list is receiving updates. |
| Investigate | Search of detections over a time window, from the last hour to the last 30 days. |
Detection needs an active organisation membership. Beyond that, the module is open to everyone in the organisation, but the server refuses a review verdict from anyone who is not an administrator or editor, and refuses adding a camera through the Add camera wizard from anyone who is not an administrator. The screen then reports the refusal.
Live: the camera wall
The wall shows 1, 2×2, 3×3 or 4×4 tiles. A toggle switches between Fixed cameras and Field live. Each tile shows a health label (Live, Connecting, Recovering, Degraded, Stale video, Offline, Not checked or Out of service), an analysis badge, and, when video is not playing, a media state. Before a session starts these are "Checking media source…", "No media source configured" (the camera has no stored connection) or "Connector not supported for live media"; the player has other states of its own, such as connecting, reconnecting, unavailable and unauthorised.
Tile controls appear when you hover over a tile: Snapshot (only while video is live), Camera details, Live preview (only when the camera can be streamed), Test connection, Detection setup, Fullscreen and Remove from wall. The toolbar adds Add camera and Discover, and three commands (Health, AI Analysis, Recording) that need a camera selected on the wall first. Adding cameras is described in Cameras and data sources.
Detection setup for a camera
Each camera has a setup dialog with tabs for capabilities, analysis, recordings, zones, rules and runtime. A rule says what should happen when a given capability fires. The outcomes offered are:
- Record only.
- Require human review.
- Create an incident automatically. Offered only for a capability that does not require human review.
- Notify. Listed, but no notification is sent in this release.
Events and review
Every event has a review state: Candidate, Review required, Confirmed, Dismissed or Escalated. The analysis pipeline can only set an event to Candidate or Review required; only a person can confirm, dismiss or escalate it.
The Events tab filters by site, severity (critical, high, medium, low, info), status (open, acknowledged, closed), capability and review state, and searches the event description. It loads the 200 most recent events and can export what is loaded as CSV. Each row shows severity, review state, a headline, the model's score and a time.
The detail panel shows Provenance, Human review and Review history, and Detected objects and Possible correlations when the event has them. The buttons depend on the event's state: Confirm, Dismiss and Escalate to incident while it is open, Acknowledge only while its status is open, only Escalate to incident once it is Confirmed, and none once it is Dismissed or Escalated. A verdict can carry an optional reason, which is recorded in the audit trail. The panel warns that a review verdict is attributed to you and cannot be undone.
Where events come from: mainly through an authenticated detection-ingest interface that the organisation's analysis pipeline calls. Frame analysis and edge sources can also write events. There is no screen for creating an event by hand, and automatic escalation applies only to events that come through the ingest interface.
From a detection to an incident
There are two routes, and they differ in who decides.
An operator escalates. Escalate to incident records the Escalated verdict and then opens the incident form, pre-filled with the title, description, type "security" and a location. Severity is not pre-filled and starts at medium, even for a critical event, so set it yourself. Submitting the form creates the incident and links it to the event. The database returns the same incident if the same event is submitted again, but once an event is Escalated the panel offers no way to reopen the form.
A rule creates one automatically. For events submitted through the ingest interface, Virtik can open an incident without a person first reading the event, in two cases:
- an event that carries no capability and has critical or high severity; and
- an event for a capability whose rule is set to "Create an incident automatically", but only if that capability does not require human review.
Weapon, licence-plate, face and smoke-or-fire capabilities require human review and are never escalated automatically. An incident created this way is marked unverified.
Investigate
Investigate searches detections over a window, or on a date you choose, and exports the results. A search returns at most 200 detections. A detection is not linked to a recorded clip, so Investigate does not play footage: it says "This detection is not linked to a recorded clip".
What it depends on
- Cameras added to a site, and for live video a media gateway. A tile says "No media source configured" when the camera has no stored connection; without a gateway the player reports its own error instead.
- An analysis pipeline to produce events. The product reports its own readiness with messages such as "Worker offline" and "No analysis worker has reported in recently. Nothing is being analysed." when none is running.
- An AI provider or detector configured for the organisation where a capability needs one. Readiness labels include "No inference backend" and "Backend not configured".