USE CASE / ELECTRICAL SUBSTATIONS

Inside the fence is never routine.

Substations are unstaffed, remote and dark, and a single person inside the yard is an event. Semaphor helps the operator watching dozens of utility sites see that person the moment they appear.

A site with one simple rule.

Most monitoring jobs are about telling normal activity from suspicious activity. A substation is different. Outside of scheduled maintenance, nobody should be inside the fence. Any person in the yard, at the control house door or climbing near the perimeter is worth an operator’s immediate attention.

The challenge is not deciding what matters. It is noticing it. Utility and contract monitoring teams watch many remote sites at once, most of them dark and empty for weeks at a time, and the one pane that changes at 2:40 a.m. is easy to miss on a crowded wall. Semaphor is AI video monitoring assistance that watches every one of those panes continuously and pushes the change to the top of the inbox.

What operators watch at a substation.

  • Perimeter fence lines. Long runs of chain-link or palisade, often with barbed wire, sometimes along roads, fields or rail lines.
  • Vehicle and pedestrian gates. The entry points crews use, and the ones intruders test or cut.
  • The switchyard. Transformers, breakers, bus work and grounding. Copper grounding and cable are the target in most thefts.
  • Control house. Door and exterior, where protection and communications equipment live.
  • Laydown and spare equipment areas. Reels of cable, spare parts and scrap that attract metal thieves.
  • Access road. A vehicle pulling off and stopping near the fence at night is often the first sign.

What matters, and what doesn’t.

Matters: a person anywhere inside the fence outside a scheduled work window. A person walking the outside of the fence line, stopping at a section, or returning to it. A vehicle stopped on the shoulder beside the site with no lights. Someone at the control house door. Activity in the laydown yard at night.

Doesn’t: deer, coyotes and livestock along the fence. Birds on the bus work and insects at the lights. Blowing brush, snow and rain. Headlights from the road sweeping the yard. Scheduled crews during their work window, which operators can clear quickly once the AI description confirms it is a marked truck and crew in hi-vis.

One person inside the fence is a Critical-level question for most utility post orders. Semaphor’s job is to make sure the operator sees that pane first.

What your guards watch

  • Dozens of remote sites that look identical and empty
  • Thermal or low-light views full of animals and weather
  • Fence lines longer than any one camera can cover well
  • Motion alarms nobody trusts anymore

What Semaphor surfaces

  • A person inside the yard, flagged on the pane where they appeared
  • A vehicle stopped beside the fence on the access road
  • An AI scene report stating what it sees and what is uncertain
  • Hashed original frames for the utility’s investigators and police

Remote-site camera problems.

Substation cameras are usually mounted high on poles covering wide yards, so a person can be small in the frame. Nights are fully dark apart from a few yard lights. Some sites use thermal or low-light cameras, others standard color with IR. Bandwidth to remote sites is often limited, so the monitoring dashboard may update slowly or show lower-quality streams, and many centers rotate through site pages because there are too many cameras to show at once.

Semaphor works with the camera software you already use, reading the same rendered dashboard the operator sees rather than connecting to each remote recorder. It keeps track of each camera as the wall rotates, and when a site’s page cycles off-screen, it marks that camera unobserved. For utilities, that honest coverage record matters: you know which sites were watched and when.

How Semaphor supports utility monitoring.

Local person and vehicle detection runs on every pane. The detector is looking for people and vehicles, not raw motion from animals and weather, and a person who stops moving is kept supported by tracking and regional rechecks instead of disappearing from the scene.

Alerts are progressive. An activity card publishes immediately, and a vision-language model then updates it with a scene report and a recommendation. Alert prioritization puts inside-the-fence events at the top, with Critical, High, Unassessed and Low levels the operator can override. The operator decides whether to call the utility’s security desk, dispatch a patrol or contact police. Semaphor never takes that decision.

Every alert keeps the original frame and crop, an exact SHA-256 hash and perceptual hashes, with an exportable JSON manifest. That evidence trail supports theft investigations and prosecutions. For overnight procedures, pair it with our guides to after-hours trespassing and alarm verification for police response.

Utilities with a central security team can run Semaphor in their GSOC; those that contract monitoring can have their remote video monitoring center run it on the operator station.

See your remote sites through Semaphor.

Book a live demo on your own camera screens, or start a pilot on one monitoring station covering a group of substations. We measure missed events, nuisance alerts and response delay on your real overnight footage.

Substation surveillance FAQ

Does Semaphor connect to our substation network or SCADA?

No. It reads the camera monitoring screen only. It has no connection to protection, control or SCADA systems.

Will animals along the fence trigger alerts?

Alerts are driven by person and vehicle detection, not raw motion, so deer, brush and snow are not what triggers a card. When the view is ambiguous, the AI scene report says so.

Can it work with thermal cameras?

Semaphor watches whatever appears on the monitoring screen. We confirm detection quality on your specific thermal views during a pilot.

BOOK A LIVE DEMO

See Semaphor on your own camera screens.

Bring the monitoring layout your guards use every shift. We’ll show focused alerts, evidence and AI context on it, then plan a pilot around one station.