Wildfire Smoke Detection AI

Add a real-time detection layer to every camera in the service territory with vision AI for wildfire smoke detection. Built for the utilities where a plume twenty miles out is a few dozen pixels against a ridgeline, the fog that rolls up the valley every autumn morning looks like exactly the thing you are watching for, and the distance between a spot fire and a campaign fire is measured in the minutes before anyone picks up the phone. Whether you're running a PTZ network across high fire-threat districts, pulling partner and ALERTWildfire feeds, or watching right-of-way after a fault, Roboflow detects smoke on the cameras your territory already runs, with a record behind every alert and every dismissal.

Early Detection Across the Camera Network:

• Detect smoke plumes at long range, where the signal is a faint column against terrain rather than an obvious cloud
• Hold detection through PTZ pans, preset tours, and the changing horizon each camera sees across a shift
• Watch for night ignition signatures, glow and thermal, when a daytime plume model has nothing to work with

False Alarms, Terrain, and Operator Trust:

• Separate smoke from the fog banks, valley inversions, dust from field work, cooling tower steam, and low cloud that trip generic motion alarms
• Learn what each camera's view looks like across seasons, so a model tuned in July still holds in November
• Track dismissals alongside alerts, so the desk watches the false-positive rate fall and keeps trusting the alarm

Response, Correlation, and Reporting:

• Correlate a detection with fault and recloser events on the same circuit, so an ignition after a fault surfaces in minutes rather than at the morning briefing
• Push alerts with camera, bearing, and imagery into your situational awareness platform and dispatch, so the confirmation call starts with a picture
• Keep a detection-to-dispatch record that supports wildfire mitigation plan filings, regulator inquiries, and after-action review

Bring intelligence to every camera today. Stop a plume on the ridge from becoming a campaign fire, a shutoff wider than the day required, or a timeline you have to defend.

‍

More About Wildfire Smoke Detection

What is wildfire smoke detection with Vision AI?

Wildfire smoke detection with vision AI uses computer vision models to watch a utility's camera network continuously and flag smoke while a fire is still small, typically minutes before a call comes in from the public. Models run against existing PTZ and fixed cameras across high fire-threat districts, hold through preset tours and changing horizons, and push an alert with the camera, the bearing, and the frame so the situational awareness desk starts a confirmation call with a picture rather than a rumor. The result is a timestamped detection and response record that supports CPUC wildfire mitigation plans, AB 1054 safety certification, and state WMP reporting.

Can Vision AI tell smoke apart from fog, dust, and low cloud?

This is the question that decides whether a smoke program succeeds, and it is worth answering directly rather than with a number. Fog banks, valley inversions, dust from field work, cooling tower steam, and low cloud all look like a plume to a generic motion or change detector, and a desk that gets buried in false alarms in October stops trusting the system by August. The approach that works is per-camera and per-season: models trained on each camera's actual view, including its specific false-positive sources, and on the seasons that view moves through, rather than one global smoke model dropped across a whole territory. The honest limits are worth stating too. Night detection is a different and harder problem than daytime plume detection, and depends on glow and thermal signatures rather than a visible column. A camera needs line of sight, so terrain shadows are real gaps. And detection is a trigger for human confirmation, not an autonomous dispatch decision; the value is in the minutes it buys the person making the call.

Does wildfire smoke detection support CPUC wildfire mitigation plan and AB 1054 requirements?

Yes. Roboflow detection can run as documented monitoring within a wildfire mitigation program, supporting CPUC wildfire mitigation plan filings in California, AB 1054 safety certification, and the equivalent state programs in Oregon, Colorado, New Mexico, and Texas, generating the detection timestamps, camera coverage evidence, response timelines, and after-action documentation those filings and inquiries run on. Roboflow is the detection engine; your wildfire mitigation, grid operations, and legal teams own the alerting thresholds, the confirmation protocol, the shutoff criteria, and the reporting. Detection supplements a mitigation program that also rests on asset condition, vegetation management, and grid hardening rather than substituting for any of them.

Can it integrate with our situational awareness platform, SCADA, and dispatch?

Yes. Detections push into wildfire situational awareness platforms, GIS, and common operating pictures through REST, MQTT, and webhooks, carrying the camera ID, bearing, timestamp, and the frame that triggered the alert. Detections can be correlated against SCADA and OMS fault, recloser, and EPSS trip events on the same circuit, so an ignition following a fault surfaces as a linked event rather than two records nobody joins until the next morning. Alerts can route to the situational awareness desk, to dispatch and CAD, and to mobile for field crews, and video can process at the edge at remote camera sites where backhaul is thin, which matters across mountain and desert terrain where bandwidth is the real constraint.

‍