Somebody upgrades their cameras. Same positions, same lenses, higher resolution. Daytime footage looks sharper, exactly as promised. Then something happens at two in the morning, they pull the clip, and the person in it is a grainy smear moving through a car park — worse than the old cameras managed on the same spot.
Nobody swapped in a defective product. This is a known trade, it is predictable before purchase, and almost nobody is told about it, because the number on the box only describes the half of the trade that sounds good.
The sensor is a fixed piece of real estate
Start with what a camera sensor actually is: a rectangle of silicon, physically small, divided into a grid of light-collecting sites. Each site becomes one pixel in the image.
The megapixel figure tells you how many sites that grid is divided into. It tells you nothing about how big the rectangle is.
That is the whole thing. Two cameras can carry sensors of exactly the same physical size, and one of them divides that area into far more pixels than the other. The rectangle did not grow. The pixels got smaller, because the only way to fit more of them into the same area is to make each one smaller.
We have written separately about what "enough resolution" actually means — how many pixels land on the thing you care about, at the distance where it happens. That article is about where your pixels go. This one is about what each pixel is worth, and the two answers pull in opposite directions.
Small pixels collect less light
A pixel is a bucket. Over the fraction of a second the shutter is open, light falls on the sensor, and each site collects whatever lands within its own boundary. A bigger bucket catches more.
That is not a manufacturing weakness anyone can engineer away with a firmware release. It is geometry. Halve the area of a collecting site and, all else equal, roughly half as much light lands in it during the same exposure.
In daylight this rarely matters. There is so much light arriving that even a small site collects plenty, and the extra pixels give you genuinely more detail. This is why the daytime demo is honest and the daytime demo is also useless as a predictor.
At night there is very little light arriving. Now the size of each bucket is the thing that decides whether you get an image or a guess, and the camera that divided its sensor into more, smaller buckets is the one working with less light per pixel.
So the same design change that improved the daytime image degraded the night one. Both are real. Neither cancels the other.
What "not enough light" actually looks like
A pixel that collects very little light does not simply produce a darker image. It produces an uncertain one.
Every image sensor carries a floor of electrical and thermal noise — random fluctuation that exists whether or not anything is being photographed. When a pixel has collected a lot of light, that random component is small compared to the real signal and you never notice it. When a pixel has collected very little, the random component is a meaningful share of what the pixel reports. The result is the speckle you recognise instantly in night footage: a grainy, crawling, colour-flecked image where flat surfaces shimmer.
That noise is not cosmetic. It is the camera telling you it does not know what is there.
Then three things happen, and each of them is the camera trying to help.
One: gain, which amplifies the uncertainty too
The camera turns up the electronic gain to make the dark image brighter. Turning up gain amplifies everything the pixel reported — the real signal and the noise together. You get a brighter picture that contains no more information than the dark one did. A face that was not captured does not become captured by being made lighter.
Two: noise reduction, which is where moving people disappear
Every camera runs noise reduction. The most effective kind compares the current frame against previous frames: if a pixel is fluctuating frame to frame but the scene is not changing, the fluctuation is noise, so average it away. On a still scene this works beautifully. A parked car, an empty loading bay, a closed door — all of it cleans up.
Now put a person in it.
The camera cannot tell the difference between "this pixel changed because of noise" and "this pixel changed because a person moved through it." Averaging across frames does to a moving person exactly what it does to noise: it blends them into what was there before and after. The person acquires a soft ghost trail, edges dissolve, and facial detail — which is fine, low-contrast structure — is precisely the kind of detail that averaging removes first.
This is the smearing you have seen and probably blamed on the recorder. The still parts of the frame look clean. The one thing you needed is a blur.
The harder the camera has to work to suppress noise, the more aggressively it smears. And the camera with smaller pixels has more noise to suppress.
Three: longer exposure, which blurs the moving subject
The other way to gather more light is to leave the shutter open longer. If each frame is exposed for a longer slice of time, more light reaches each pixel and the image genuinely improves — for anything that stays still.
Anything moving is now recorded across the whole of that longer exposure, and comes out stretched. A walking person's face smears in the direction of travel. A hand becomes a streak. A licence plate on a moving vehicle becomes an unreadable band.
Read that sequence again, because it is the trap in one line: the camera compensates for low light by increasing exposure, and increased exposure destroys detail on moving subjects. Which means the failure arrives exactly when someone is moving, at night — the event you bought the camera for. Static, well-lit scenes are the one condition where none of this bites.
Sensor size: the number nobody asks for
Here is the question that would have exposed all of this at the quoting stage, and it is almost never asked: how big is the sensor?
Buyers ask for megapixels. Sensor size is what makes megapixels mean something. The same pixel count on a physically larger sensor gives you larger collecting sites and a better night image, and larger sensors cost more to manufacture, need larger lenses, and make bigger camera bodies. That cost is real, so it shows up in price without showing up in the headline specification.
Two cameras quoted at the same resolution can be materially different products at night for this reason alone, and nothing on the datasheet line you were shown distinguishes them.
You do not have to master the notation sensor sizes are published in. You have to ask for the figure, ask how it compares to the alternative being quoted, and notice whether the person quoting can answer at all.
This is a trade, not an argument for buying less
We want to be exact here, because this argument is easy to overshoot into nonsense.
More resolution is not a defect. On the same sensor size, more pixels genuinely gives you more detail in good light, and detail is what identification is made of. A camera that cannot resolve a face in daylight will not resolve one at night either. Nobody should read this article and go buy the lowest-resolution camera available.
The claim is narrower and it is this: on a given sensor size, resolution and low-light capability trade against each other, and buying purely on the megapixel number means making that trade blind, in the direction that hurts you when it counts.
The right resolution is the lowest one that satisfies the identification task at that specific spot, at the distance that matters — because every pixel beyond that is being paid for twice. Once in money, and once in night performance.
That is why the two articles fit together. Work out the pixel density the spot actually requires. Then choose the camera that meets it, on the largest sensor available in budget, rather than the camera that maximizes a number.
The parts a sensor cannot fix
Two honest limits on everything above.
Light is still the cheapest fix available. A modest amount of well-aimed lighting does more for night footage than a sensor upgrade, because it changes the amount of light arriving rather than fighting over what to do with too little. This is unglamorous, it usually costs less than the camera it replaces, and it also has a deterrent effect a camera does not. Infrared illumination is the alternative where visible light is unwanted — with the trade that the image goes monochrome, and that illuminators light what is near them, not what is far away.
Thermal cameras solve a different problem, not this one. Thermal imaging does not depend on visible light at all — Axis states that "since objects themselves emit the heat that thermal cameras detect, thermal cameras are not dependent on visible light and can detect in all light conditions." That makes it excellent for knowing something is out there in total darkness. It does not replace an identification camera, and Axis says so plainly: "unlike conventional cameras, this technology only allows for detection", and "while it can determine a person's shape and size, it doesn't allow for clear identification." Thermal answers is someone there. It does not answer who.
What to ask before you buy
Four questions. Any competent designer can answer all four, and the answers are more useful than any specification sheet.
- What size is the sensor in this camera, and in the alternative? If the answer is only a megapixel figure, the question has not been answered.
- What is this camera's exposure behaviour at night — does it hold a fast enough shutter to freeze a walking person, and what does it give up to do that? Every camera has a setting that trades brightness against motion blur. Ask which way it is set, and ask for it to be set deliberately rather than left on default.
- Do I need identification at this spot, or detection? They are different jobs with different requirements. Both are legitimate. Paying identification prices for a spot that only needs detection is where budget quietly disappears — and the seven-level scale published against IEC 62676-4 shows how far apart those requirements sit.
- Can you show me night footage of a person walking, at this spot, from this camera? Not a brochure image, not a static night scene. A person, moving, at your site. This one question tests everything in this article at once, and it is the only test that cannot be answered with a specification.
The uncomfortable version of all this: the upgrade that improves your daytime footage may be the one that costs you the night. Ask which one you are buying.
