Skip to main content

Thinking Mode

Thinking mode interprets risk events such as fire, smoke, and fall-down incidents.

Best Use Cases​

  • When incident-level interpretation is more important than simple object detection
  • When operations are centered on risk-event alerts


Example Inputs​

  • Alert when smoke or fire appears
  • Alert when a person falls down

In Thinking mode, alert context is shown as Incident detected.



Template Tips​

  • Standardize common safety incident types (fire, smoke, fall) as reusable templates.
  • If risk definitions differ by site, include site names in template titles.
  • Avoid combining too many risk types into one scenario; split by incident type.


Writing Tips​

Thinking details are usually centered on the incidents block.

Operational tips:

  • Prioritize incidents that can be clearly judged in a single scene
  • Avoid overly complex phrasing; keep core risk statements short
  • Start from frequently observed risk types in the field

Alert message examples:

  • Incident detected: smoke or flame
  • Incident detected: person collapsed on the floor


Operational Cautions​

Thinking mode works best for immediately visible risks.

  • Prefer clearly visible incident types over scenarios requiring long-context behavior tracking.
  • In environments with reflections, haze, or occlusion, check camera position and lighting before tuning scenarios.
  • Ambiguous scenes may be judged conservatively, so validate with representative footage during initial rollout.


Multi-frame Usage​

If temporal context matters, adjust frame mode.

  • Sequential flow: sequence
  • Time-window analysis: timeline

Example:

frame_mode:"sequence",
frame_mode     : "timeline"
frame_interval : 10

timeline is effective when explicit time conditions are required.



Technical Blog​

For deeper technical background, see Thinking mode for more accurate visible risk interpretation.