Some questions do not deserve a good camera. Is the tray flooded. Is the cooler door shut. Is the pump running. Did the vent open. Is something moving in the barn at 2am. Those are yes-or-no questions, mediocre pixels answer them completely, and the interesting number is not resolution; it is how many places you can afford to watch. Twenty $8 cameras across a site is a different capability from three good ones, and that is the capability this config buys.
The need.
What you want to know: is that thing over there the way I left it?
Every operation accumulates spots that only get checked when someone happens to walk past: the far float valve, the propagation tray at the end, the pen gate, the pump house. Each one is a five-second glance that costs a fifteen-minute walk. A witness camera makes the glance free, from anywhere, on every one of them at once.
The solution.
Each station shoots a still on a schedule (and can shoot extra on motion, the classic trail-camera pattern) and pushes it to your endpoint. The record is a strip of glances: this morning's tray, last night's pen, the door at closing time. Your endpoint is where "the 6am frame looks wrong" becomes an alert; the witness just testifies.
What this tier is not.
It is tempting to point an $8 camera at plants and call the money saved. Resist it, with the reasons in hand: the deployed cost of any station is the enclosure, power, mount, and labor, call it $60 to $120, so the cheap board saves perhaps $30 on the one part that decides the data quality for the station's whole life. And its dated 2 megapixel sensor, with automatic processing you cannot fully turn off, loses exactly what plant work needs: color fidelity and resolution. Where a view has power and the question is the crop, spend the $30 and build the Plant Station. Where the question is yes-or-no, build ten of these and watch everything. The full argument, including where the ESP32 genuinely wins outright, is on the imaging page.
The config: the parts.
| Part | Job | Cost |
|---|---|---|
| ESP32-CAM board (with its flashing adapter) | Camera and Wi-Fi in one module. Fiddly to flash the first time; cheap enough to deploy in numbers. | ~$8–12 |
| A 5V power supply per station | The known weak point: brownouts on thin supplies cause most ESP32-CAM grief. Use a solid one. | $5–10 |
| A small enclosure and mount per station | A box, a zip tie, a bracket. The witness tier is allowed to be scrappy. | $5–15 |
| A Station Card per station (optional here) | Identity in the frame. Worth it once you have more than a few witnesses. | A few dollars |
The ESP32-CAM hardware page covers the board's quirks, including the flashing adapter and the brownout habit. A newer ESP32-S3 camera board costs a few dollars more and removes most of the fiddliness; either fits this config.
The Station Card here.
Honestly: optional, and here is the honest reasoning. A witness answers yes-or-no, and yes-or-no does not need color correction or a scale bar. What the card still earns at this tier is identity: with fifteen near-identical frames of near-identical corners arriving at one endpoint, a card in each view names the station inside the picture itself, which survives every renamed file and reshuffled folder. Fit the card where it is easy; skip it where it is not; the one non-negotiable it carries elsewhere (color you can trust) is not a promise this tier makes anyway.
The sketch.
Coming
The witness sketch (timed stills, optional motion wake, push to your endpoint, the same captive-portal setup and browser flashing as the rest of the sketch library) is being written and is not downloadable yet. This page is the spec it is being built to. The parts list is stable, and the board itself is the same one the community has run for years.
Frequently asked questions.
What is the cheapest way to put a camera on a barn, coop, or pump house?
An ESP32-CAM board, around $8 to $12 with its flashing adapter, plus a solid 5V supply and a small enclosure: roughly $20 to $35 per spot, deployed. It shoots a still on a schedule and pushes it to a place you own, no subscription and no cloud account. The picture quality is modest, which is fine, because the job is seeing state, not detail.
Is the ESP32-CAM good enough for monitoring plants?
For watching a place (is the tray flooded, did the vent open), yes. For measuring plants, no: its 2 MP sensor and uncontrollable automatic processing lose the color fidelity and resolution plant work depends on, and the board is only a small fraction of a deployed station's real cost. Where the question is the crop and power exists, a Raspberry Pi camera station is the honest tool for about $30 more.
Can a cheap Wi-Fi camera work as a trail camera for predators?
Yes, within Wi-Fi range: pair the camera board with a motion sensor so it sleeps on batteries and wakes to shoot whatever moved, the classic trail-camera pattern built open, with the pictures landing at your endpoint instead of a subscription app. Beyond Wi-Fi range, a commercial cellular trail camera is the practical tool.