It is one of the most effective lines in this industry, and the reason it works is that it is not a lie. One camera, one cable, one licence, one hole in your ceiling, and the whole room is covered. Against four cameras with four cable runs and four brackets, the arithmetic looks decisive before anybody has looked at the room.
The honest answer is that a fisheye replaces four cameras for one job and fails to replace even one of them for another. Which job you are buying is the entire question, and it is almost never the question that gets asked.
What the lens is actually doing
A conventional camera lens takes a modest slice of the world — perhaps a doorway and the floor in front of it — and projects it across the whole sensor. A fisheye lens takes a hemisphere and projects that onto the same sensor.
The result, before any software touches it, is a circular image inside a rectangular frame. Mounted on a ceiling looking straight down, the centre of that circle is the floor directly beneath the camera and the outer edge of the circle is the horizon in every direction at once. Everything in the room is somewhere on that disc.
That image is unreadable to a person, so the camera or the recording software dewarps it: it takes regions of the circular image and mathematically re-projects them into rectangular views that look like normal camera footage. Most systems give you several dewarped views at once from the single stream — a "quad" that mimics four cameras, a panorama strip, or a virtual pan-tilt-zoom view you can drag around inside recorded footage after the fact.
That last capability is genuinely useful. Because the whole hemisphere was recorded, you can go back to last Tuesday and look anywhere in the room, including at things nobody thought to point a camera at. A conventional camera only ever recorded where it was aimed.
But here is the part the demo does not say out loud. Dewarping creates views. It does not create pixels. Every dewarped view is a crop of the same original circular image, stretched back into a rectangle. Four views from one fisheye are four slices of one sensor. Four cameras are four sensors.
The pixel-budget trade
We have written elsewhere about pixel density — the pixels-per-metre your camera actually delivers at the spot where an incident happens, as opposed to the megapixel count on the box. That article is the foundation for this one. If you have not read it, the short version is that a sensor is a fixed budget of pixels spread across whatever the lens is pointed at, and the share landing on a person is what decides whether you can say who they were.
A fisheye is the most extreme version of that trade available for purchase. It spreads the entire budget across a hemisphere.
Here is the arithmetic, framed honestly as a worked example with assumed inputs — these are not measurements from your building and not figures from any standard:
Assume a 12-megapixel fisheye whose circular image is about 3000 pixels across, mounted centrally on a ceiling.
- If that circle covers a floor area roughly 6 m across, the average works out to about 500 pixels per metre.
- If the same camera is asked to cover a floor area roughly 20 m across, the same 3000 pixels are spread over 20 m: about 150 pixels per metre.
- At 30 m across, about 100 pixels per metre.
One camera. Same sensor. The only thing that changed is how much floor you asked it to cover, and the useful density fell by a factor of five.
Now compare that against the alternative. Four conventional cameras covering the same 20 m room each get their own full pixel budget aimed at a quarter of the space. The fisheye did not replace four cameras there. It replaced four cameras' worth of coverage with one camera's worth of detail.
Two further things make the real picture worse than that even arithmetic suggests, and both are geometry rather than opinion:
The outer ring is compressed. A fisheye maps angles onto the image roughly evenly, but out near the edge of the circle each degree of angle covers a great deal more floor than a degree near the centre does. Pixel density therefore does not fall gently as you move outward from beneath the camera — it falls away sharply. The person standing directly under the camera is well resolved. The person at the far wall is not, and the difference is much larger than the difference in their distances.
Dewarping stretches. Straightening a curved slice back into a rectangle interpolates. The output looks like clean rectangular footage, which is exactly the problem — it looks like a normal camera view, so people trust it like one, while carrying the detail of the crop it came from.
Ceiling height changes the answer
This is the dependency that decides most real installations, and it cuts both ways.
A ceiling-mounted fisheye sees a circle whose size is set by how high the camera is. Low ceiling, small circle. High ceiling, large circle. So a retail floor with a 3 m ceiling and a warehouse with a 9 m ceiling are not the same problem with different budgets — they are different problems.
In a low room, the camera covers a modest area, which is the good case: the pixel budget is concentrated and the density stays useful across most of the view. The trade-off is that you may need more than one unit to cover a long space, at which point some of the cost advantage evaporates.
In a high space, one unit covers an impressive area — and this is where the one-camera-replaces-four claim is most enthusiastically made and least true. The same pixels are spread over far more floor, and everything at the outside of the circle is a shape. Worse, from that height the camera looks steeply down at everyone. It sees the tops of heads and shoulders, and where people went. It does not see faces, and no dewarping recovers a face the lens never captured.
That geometry is the same reason we mount face-capture cameras at head height rather than up under the eaves. A ceiling fisheye is, by design, the opposite of a face-capture camera.
Where a fisheye genuinely wins
There is a real category here and it deserves to be stated as plainly as the limits.
A fisheye is the right choice when your question is "what happened?" rather than "who was that?", over a small to medium space, from a normal ceiling height:
- A small retail floor. One unit over the sales area gives you continuous coverage of every aisle with no blind spots between camera fields — and the aisle gaps between conventional cameras are where a surprising amount of loss actually occurs.
- A lobby or reception area. Who came in, where they went, how long they waited, which door they used.
- A boardroom or open-plan office. Meeting-room disputes, after-hours access, equipment that walked.
- A back-of-house area, a till zone, a stockroom. Sequence-of-events coverage of a room you rarely need to identify a stranger in, because the people in it are mostly people you know.
Reconstructing movement is the fisheye's real, under-sold advantage. Four conventional cameras produce four clips you have to mentally stitch together. A fisheye produces one clip in which the whole event is visible at once. For incident reconstruction — for insurance, for a staff conversation, for understanding your own operation — that is a better artifact.
Where it fails, and should not be sold
- Large warehouses and open floor plates. The area is too big and the ceiling too high. You will get a beautiful overview and nothing usable at the perimeter.
- Long approaches, driveways, yards and parking areas. These need reach. A hemisphere is the opposite of reach.
- Any identification task at distance. If you need to establish who a person was, at a specific spot, a fisheye covering the whole room is the wrong instrument for that spot. Full stop.
- Licence plates. Not a candidate. Plates need high density on a small, fast-moving target.
- Rooms that are long and narrow. A circle fits a square room. In a corridor-shaped space, most of the circle lands on walls.
There is one more failure mode that is commercial rather than optical: the recorder and the licensing. A high-resolution fisheye producing multiple dewarped streams asks more of your recording infrastructure and, on some platforms, is licensed as multiple channels rather than one. If the pitch was "one camera, one licence", ask for that in writing before it becomes a surprise.
The honest hybrid answer
The design that actually works on most sites is not one or the other. It is:
A fisheye for the overview, plus a conventional fixed camera at each choke point.
The fisheye owns the room. It tells you the sequence — how many people, where they went, what was touched, how long it took. The fixed camera owns the doorway, the till, the gate, the top of the stairs — the small number of spots where you need to identify a person rather than observe one. It is a narrow view at a useful angle, and it carries the pixel density to do the identification job.
That is usually two or three cameras where the fisheye pitch promised one and the conventional design proposed six. It costs more than the first and less than the second, and it is the only one of the three that answers both questions.
IEC 62676-4 (current edition 2025) is the application-guidelines standard manufacturers work from when they publish the pixel densities required for tasks ranging from bare overview through to establishing identity. We have not bought a copy and will not tell you what is inside it. The point here is structural and does not depend on reading it: overview and identification sit at opposite ends of that range, separated by roughly an order of magnitude in pixel density, and no single camera delivers both ends across a large space. That is not a product limitation anyone can engineer around. It is a fixed budget being divided.
What to ask before you buy one
- What floor diameter will this cover at my ceiling height, and what pixel density does that give at the far edge — not the centre? A vendor who designs properly answers this before quoting.
- Which specific spots do I need identification at? Then check whether the fisheye reaches them. Usually it does not, and that is fine — it just means one more camera, not a different overview.
- Am I being shown a dewarped view from the centre of the image? Ask to see the far edge instead. Ask to see it at night, with a person in it, in your building.
- How is this licensed on the recorder, and how many streams is it producing? One camera is not always one channel.
- What is this camera's job in one sentence? If the answer is "everything", the design has not been done yet.
The fisheye is a good product that is frequently sold for the wrong reason. Bought as a room-awareness camera it is excellent value and often better than the four cameras it displaced. Bought as a way to avoid buying the camera that watches your front door, it is the most expensive saving on the quote.
