Cable theft detection only matters if it fires before the cut. Here is how to set that up on the cameras you already own.

1. Why Most Cable Theft Systems Fire Too Late

The standard cable theft "detection" stack is a power monitoring relay plus a motion-triggered camera. The relay only trips once cable continuity is broken, which means the wire is already cut. The motion camera triggers on movement, but at 2 AM on a Saturday at a transformer yard, motion alerts go to a recording, not a person. The chain of events looks like this: thief arrives, thief approaches the cable run, thief cuts, power drops, relay alerts a tech who may take 30 to 90 minutes to respond, thief leaves with the copper before that response reaches the site.

A camera pointed at the transformer also records all of this. The footage is excellent for an insurance claim. It rarely leads to recovery. Copper is fungible and untraceable once it hits a scrap yard, and most jurisdictions do not prioritize copper theft investigations unless the dollar value crosses a felony threshold per single incident.

The unit economics are brutal. A single transformer or service feed pull can cost $15,000 to $80,000 in copper replacement, plus utility downtime, plus the labor to re-pull and re-terminate. The thief gets $200 to $800 for the same copper at the scrap yard. The asymmetry means thieves keep coming back as long as the only consequence is being recorded.

2. The Pre-Cut Window: What 90 Seconds Buys You

Watch any surveillance footage of a copper theft and time the segment between "vehicle stops" and "tools touch wire." It is almost always between 60 and 180 seconds. The thief parks, scans the area, opens a tool bag or trunk, walks to the conduit, and only then makes the first cut. That window is the entire ballgame. If a live audio warning broadcasts during those 90 seconds, the thief leaves. If a police dispatch goes out at second 30, the patrol car is already rolling before the cut starts.

Conventional motion alarms fail this window because they are not selective. A raccoon, a piece of trash blowing across the lot, a delivery driver turning around: all of these trigger the same alert as a thief, so the alerts get muted or auto-dismissed. The system has to be specific enough to send a human-grade alert only when a human is approaching a sensitive zone, and it has to do that in seconds, not minutes.

3. Zone-Based AI Detection on Existing Cameras

Zone detection means you draw a polygon on the camera image around the thing you actually care about (the transformer pad, the conduit run between buildings, the door to the electrical room) and the AI only alerts when a person crosses into that polygon. Movement outside the zone is ignored. A delivery truck driving past the parking lot does not trigger anything. A person stepping over the curb toward the transformer does.

The classifier matters. Generic motion detection fires on anything. A trained person detector ignores wildlife, weather, and vehicles, and only alerts on humans. Cyrano runs this classification on every frame of every connected camera, and the alert that goes out includes a snapshot from the triggering frame, the camera label, and the zone name. The recipient sees within one glance whether this is a worker who arrived early or a stranger walking toward a transformer at 3 AM.

This works on cameras you already own because Cyrano connects to the existing DVR or NVR over ONVIF or RTSP. There is no camera replacement, no new cabling, and no monthly fee per camera. You point the existing fixed cameras at the things you want protected, draw the zones in software, and the detection runs.

4. Edge Processing vs Cloud (and Why Thieves Cut the Internet)

Sophisticated cable theft crews cut the internet feed before they touch the copper. Cloud-only camera systems go dark the moment the WAN drops, and any "detection" that depends on uploading frames to a server stops detecting. Edge processing solves this: the AI runs on a small device on-site, alerts queue locally, and when connectivity returns the alert delivers with the snapshot intact.

Cyrano ships as a $450 one-time hardware unit that sits next to the DVR and processes video locally. If the unit is on a small UPS (a $60 add-on), it survives a power cut as well. The detection still fires, the alert still gets stored, and as soon as cellular failover or the next available link comes up the alert ships out. This is the difference between a system that works the night you actually need it and a system that fails exactly when the threat is real.

The anchor here is specific and verifiable: edge inference, ONVIF/RTSP video pull from any compliant DVR, local snapshot generation, queued alert delivery. None of this depends on the camera vendor's cloud being up, which is the failure mode that turns most "smart" camera systems into expensive blind spots during a real incident.

5. Where to Point Cameras for Cable Theft Specifically

Cable theft is not random. The targets are predictable, which means camera placement should be too. The five highest-value targets on a typical commercial or multifamily site are: the utility transformer pad, the main electrical service entrance, HVAC condenser units (copper line sets and grounding wire), exterior conduit runs between buildings, and the telecom or fiber demarc. On construction sites, add temporary power poles, spools of unspooled wire, and the gang box.

Each of these benefits from a dedicated camera angle that frames the asset tightly, so the detection zone can be drawn close to the thing being protected. A wide-angle camera covering "the back of the property" cannot tell the difference between a person walking near the dumpster and a person walking up to the transformer. Two cameras with focused fields of view will outperform one camera with a panoramic view every time, both for detection accuracy and for the quality of the snapshot that goes out with the alert.

If you cannot move cameras, you can still draw a tight zone inside a wider field of view and the detection will respect it. The trade-off is that the snapshot in the alert will be lower-resolution for the triggering region. Worth doing as a starting point, but plan to add a tight camera at any asset that gets hit more than once.

6. Response Protocol That Matches the Timeline

A 90-second pre-cut window is wasted if the alert lands in an inbox that nobody checks until morning. The response protocol has to match the detection speed. The two responses that actually work in this window are live audio talkdown and immediate police dispatch.

Live audio talkdown means a real operator speaks through an on-site speaker the moment the alert fires: "You in the gray hoodie near the transformer, you are being recorded and police have been called." This works because copper crews are professionals doing a risk calculation, and the talkdown changes the math instantly. The vast majority leave within seconds. Cyrano supports this through monitoring partners that watch the alert stream and trigger talkdown from existing PA systems or speakers wired to the DVR.

Immediate police dispatch is the second leg. Because the alert includes a snapshot of a person on a transformer at 3 AM, the dispatch is verified, not a generic motion alarm, which means many jurisdictions will treat it as priority 1 instead of routing it to a non-emergency queue. The combination of talkdown plus verified dispatch turns the 90-second window into a deterrent that actually fires.

7. Frequently Asked Questions

Will this work with the cheap PoE cameras already on my site?

Yes, as long as the DVR or NVR they connect to speaks ONVIF or RTSP, which covers essentially all mainstream brands (Hikvision, Dahua, Lorex, Reolink, Amcrest, Uniview, and most rebrands). Cyrano pulls video from the recorder, not from the cameras directly, so individual camera firmware does not matter.

What happens if the thieves cut the internet line first?

The on-site Cyrano unit keeps detecting and queues the alert locally. If there is a cellular failover or the link returns, the alert ships with the snapshot intact. If the unit is on a UPS, it also survives a deliberate power cut at the meter. This is specifically the failure mode that takes cloud-only camera systems offline during real incidents.

How many false alerts will I get from animals or weather?

Cyrano uses a person classifier rather than generic motion, so wildlife, blowing debris, and headlights do not trigger alerts. Combined with zone restriction (alerts only fire when a person enters a drawn polygon), real-world deployments typically see a handful of true alerts per week rather than dozens of nuisance triggers per night, which is what makes the alert stream credible enough to act on.

Is detection different for transformer pads vs HVAC condensers vs interior mech rooms?

The detection itself is the same person-in-zone classification. What changes is the camera placement and the zone shape. Transformer pads and HVAC pads need a tight zone around the equipment with the camera framed close. Interior mech rooms work best with a camera covering the door from the inside, so the alert fires when someone enters rather than when they are already at the panel.

What does this cost compared to a guard or a wired alarm?

A roving guard service for overnight coverage typically runs $3,000 to $5,000 per month. A monitored intrusion alarm on the electrical room is $50 to $150 per month but only fires once the door is breached, not at the approach. Cyrano is approximately $200 per month plus a one-time $450 hardware cost, and it covers every camera on the recorder rather than a single zone.

How fast can this be deployed at a site that just got hit?

The hardware ships overnight. Plugging it into an existing DVR and getting the first camera streaming into the detection takes under two minutes. Drawing the first zones takes another five. A site that was hit on Saturday can have pre-cut detection running before the next weekend.