The Smoke & Fire rule detects smoke and fire in a scene.

When smoke and/or fire is detected within a scene that uses the smoke & fire rule, an event will be generated, alerting operators about the incident.

When a Smoke & Fire rule is set up on a camera, every frame of the video clip recorded will be processed and reviewed to determine if there was any smoke or fire detected in any of the frames.

For the detector to pick up smoke or fire, the frame must have a proper smoke or fire present. Generally, a frame should typically have:

For Smoke & Fire detection, we recommend cameras with these capabilities

The Smoke & Fire detector has primary been trained on scenes taken from the following:

Two per-rule controls affect how aggressively and how often detection runs.

Detection sensitivity — adjusts the per-class detection-confidence thresholds. Levels: High / Normal / Low (extendable to a 1–10 scale). The higher sensitivity, the more events will be generated.

Sampling time / detection cadence — controls how often the detector samples frames of the stream. Levels map to: Fast = 2000 ms, Normal = 4000 ms, Slow = 8000 ms

Dwell time sets how long smoke or fire must persist in the scene before an event is generated. It controls how many events you receive for a given incident.

Two per-rule settings control how aggressively and how often detection runs. In most deployments the defaults give the best balance and do not need to be changed.

These settings require additional processing and are hidden by default in the Cloud portal. They are available in the on-premises interface.

Detection sensitivity — adjusts how readily a detection triggers an event. Levels: High / Normal / Low. Higher sensitivity detects more subtle smoke and fire and generates more events, but may also increase false alarms. Lower sensitivity is more conservative.

Detection cadence (sampling time) — controls how often the detector samples the stream:

Faster sampling makes detection more responsive but increases processing load on the system. Slower sampling is lighter but may take longer to pick up short-lived events. Choose the cadence that matches how quickly you need to be alerted.

Synthetic smoke (from smoke machines or cannons) and controlled fire tests look different from real smoke and fire, so detection can be less consistent than in real incidents.

Typically, smoke machines and cannons have a controlled duration of smoke and fire generation, which doesn't reflect how smoke and fire work in nature.

If you are validating a setup with test smoke or test flames and detection(s) are not triggering, try raising the detection sensitivity for the test to ensure that you are

For brief smoke or flame events, use a shorter dwell time so the incident persists long enough relative to the dwell window to raise an event. Make sure the smoke or flame still fills enough of the frame during the event.

A sustained fire will typically generate repeated events for the duration of the incident. Use a longer dwell time if you want fewer, consolidated alerts, or a shorter dwell time if you want more frequent updates while the fire is active.

We strongly recommend leaving detection sensitivity and cadence at their default values. They have been tuned across extensive real-world data to give the best balance of detection and false alarms for most environments. If you do need to adjust them: