About this skill

Problem

Real fire signals in video surveillance are often confused by lighting changes, steam, glare, or occlusion. Manual review can miss early smoke, and by the time open flames appear, the response window is short.

How the skill works

This skill analyzes an image or video for fire risk, focusing on:
- open flames and early-stage smoke
- detected anomaly count and location
- confidence and fire severity
- actionable response guidance

It uses color features, dynamic texture analysis, and smoke diffusion modeling to identify abnormal regions while reducing false positives from lighting shifts or steam. The script accepts local files via --input or remote addresses via --url, and can list cloud-side historical reports with --list. open-id is required to save and query reports, so it must be provided explicitly rather than guessed.

Boundaries

Supported inputs include jpg, png, mp4, avi, and mov, up to 100MB. The camera view should cover key fire-risk areas and avoid heavy occlusion; night infrared footage should still show fire-source contours. Output is for security alerting reference only. If a fire is confirmed, contact professional emergency services for on-site response.

Screenshots

Use Cases

A security reviewer checks a warehouse video frame for early smoke and gets response guidance.
A data-center ops engineer inspects an equipment-area clip for flame or smoke anomaly and severity.
A forest patrol analyst reviews a night IR image to confirm whether visible smoke is a fire.
A campus manager lists recent reports by area and time to locate high-risk points and recheck them.

Best For

Campus security lead: classify key camera frames as fire, no fire, or risk level.
Warehouse duty operator: upload a suspect clip and confirm smoke location plus response guidance.
Forest fire technician: verify night IR smoke traces with fewer false positives or misses.
IT facilities admin: check data-center or electrical-area video for early fire and review reports.