Cameras on sites with no power and no network: what actually breaks

A temporary deployment gets installed on a bright day by someone who checks that the picture is up, and it works. The failures arrive later: the first stretch of overcast weather, the first week the site gets busy, the first cold snap. All of them are predictable, and all of them are cheaper to design around than to diagnose from two hundred miles away.

Power: the panel is rarely the problem

Solar sizing goes wrong because the calculation gets done against average conditions rather than the worst case the site will actually see. The number that matters is not annual average sun — it is the shortest run of usable daylight in the darkest month, at that latitude, with whatever shading exists on that site.

• Size the battery for the worst run of bad days, not the average day. Three consecutive overcast days in December is a normal event, not an edge case.
• Lead-acid loses substantial usable capacity in the cold. A bank that carries the load comfortably in September can fall short at the same load in January without anything having failed.
• Shading moves. The pole that had clear sky in March gets shaded by a structure that went up in June — this is routine on construction sites specifically.
• The illuminator is often the largest single load, and it runs exactly when the panel is producing nothing.
• Heaters and blowers in the housing draw more than people expect, and they also run at night and in winter.

Connectivity: signal is a function of where you put the antenna

Cellular problems are usually mounting problems. A phone showing four bars at ground level tells you very little about what a modem will see at the top of a pole, behind a metal enclosure, next to a generator.

Bandwidth: send the decision, not the footage

Continuously streaming full-resolution video off a cellular link is the fastest way to exhaust a data plan and the least useful thing to do with the connection. The workable pattern is to do the analysis at the edge and send only what carries information: the event, the metadata, and a short clip as evidence. Full footage stays local and is pulled on demand for the small number of events anyone actually reviews.

This is the constraint that drives Rapid Onsite Analytics — a deployment that runs on solar and cellular has to make its decisions on site, because the link is not there to make them anywhere else.

The week-three failures

1. Mounting drift. Wind and vibration move a pole-mounted camera a few degrees. The detection zone that was drawn around the gate is now drawn around the fence beside it. Re-check framing against the original reference image after the first fortnight.
2. The lens gets dirty. Dust on an active site, salt near the coast, agricultural spray. Detection quality degrades gradually enough that nobody notices until it is bad.
3. Something gets parked in front of it. A container, a stack of materials, a plant trailer. This is the most common single cause of a camera that has stopped being useful on a construction site.
4. The site changed. The layout moved, the gate moved, and the camera is still watching where the gate used to be. This is not a fault — it is the nature of a site that changes every week, and it means someone has to revisit placement as the work moves.