Pixel density: what "enough resolution" actually means

Something happened, you pulled the footage, and there was a person on the screen. You could see them arrive, do the thing, and leave. What you could not do was say who it was. Not to your insurer, not to your staff, not to yourself.

You paid for 4K. The quote said 4K. And the footage is still useless for the one job you bought it for.

This is not usually a fault, and it is not usually a lie. It is a measurement nobody did before the cameras went up.

Megapixels describe the sensor. They say nothing about your scene.

A camera's resolution tells you how many pixels it produces in total. It tells you nothing about where those pixels land.

Think of the sensor as a fixed budget of pixels that gets spread across whatever the lens is pointed at. Spread that budget across a whole parking lot and each person in the frame gets a tiny share of it. Spread the same budget across a single doorway and each person gets an enormous share.

That share is the number that decides whether you can identify someone. The industry term is pixel density — pixels per metre (or per foot) across the scene, measured at the distance where the thing you care about actually happens.

This is why a well-placed 2 MP camera routinely beats a badly-placed 4K one. Not sometimes. Routinely.

The arithmetic, which you can do yourself

Here is the whole method. It is one division.

Pixels per metre = the camera's horizontal pixel count ÷ how wide the scene is, in metres, at the distance that matters.

A 4K camera has 3840 pixels across. A 1080p ("2 MP") camera has 1920.

Now run it on real situations:

• 4K camera watching a 30 m wide yard. 3840 ÷ 30 = 128 px/m. You will see that someone is there, and roughly what they are doing. You will not identify them.
• 4K camera watching a 15 m wide loading bay. 3840 ÷ 15 = 256 px/m. Now you are in useful territory for describing a person.
• 1080p camera watching a 3 m wide door. 1920 ÷ 3 = 640 px/m. Far better than either 4K example above, from a camera with a quarter of the pixels.
• 1080p camera watching that same 30 m yard. 1920 ÷ 30 = 64 px/m. A shape. A hoodie. Nothing more.

Read those four lines again, because they contain the entire argument. The 1080p camera on the door outperforms the 4K camera on the yard by a factor of five — for identification purposes — because it is not trying to cover the whole world at once.

To do this for your own site you need three things, and none of them require a technician:

1. The distance that matters. Not the furthest the camera can see. The spot where an incident actually occurs — the till, the gate, the bottom of the stairwell, the side of the trailer.
2. How wide the view is at that distance. Stand there with a tape measure, or pace it out. Left edge of frame to right edge of frame.
3. What you need to achieve there. "Know somebody walked past" and "be able to say that is definitely the person who works Tuesdays" are wildly different requirements and cost wildly different amounts.

Then divide. If the answer is smaller than the requirement in the next section, no firmware update, no digital zoom, and no upgrade to the recorder will save it. Only a narrower view, a closer camera, or a longer lens will.

What the numbers need to be

The pixel density figures that circulate in this industry are published by manufacturers as being based on the international standard for video surveillance application guidelines. That standard is paywalled and we have not read it, so we will not tell you what is in it. What we can do is point you at a manufacturer that publishes its figures openly.

Axis Communications publishes a white paper called Pixel density based on IEC 62676-4:2025 (April 2026). Working from an assumed human face width of 16 cm, Axis publishes these figures:

| What you can do | Pixel density Axis publishes |
|---|---|
| Overview — tell whether an object is moving | 3 px/face, 20 px/m |
| Outline — outline an object, see its direction | 6 px/face, 40 px/m |
| Discern — tell a person from a vehicle from an animal | 12 px/face, 80 px/m |
| Perceive — see people and movement, no characteristics | 20 px/face, 125 px/m |
| Characterize — identify by person type, gait and behaviour | 40 px/face, 250 px/m |
| Validate — verify a known person, trace actions, read a plate | 80 px/face, 500 px/m |
| Scrutinize — establish identity with high certainty | 240 px/face, 1500 px/m |

Put that against the arithmetic above and the disappointment explains itself. The 4K camera on the 30 m yard, at 128 px/m, lands at "perceive" — you can see a person moving. Being able to verify a known person is roughly four times that density. Nobody was lying to you. Nobody measured, either.

You may also have seen the older four-level scale — detect, observe, recognize, identify, sometimes abbreviated DORI. Axis's white paper written against the 2014 edition, Pixel density based on IEC 62676-4:2014, publishes those as 25 px/m (8 px/ft) for detection, 63 px/m (20 px/ft) for observation, 125 px/m (40 px/ft) for recognition and 250 px/m (80 px/ft) for identification.

The part almost every article online gets wrong

Nearly every "DORI explained" page you will find cites IEC 62676-4:2014, Edition 1.0 of Video surveillance systems for use in security applications — Part 4: Application guidelines.

That edition was withdrawn on 2025-10-09. On the same date, Edition 2.0 — IEC 62676-4:2025, same title — became the current edition.

We want to be precise about what we are and are not claiming here, because the standard sits behind a paywall and we have not bought a copy. What we have verified is that both editions exist, that they carry that title, that the 2014 edition is withdrawn, and that the 2025 edition is current. We are not going to quote clauses at you or tell you what changed inside the document, because we do not know.

What we can point at is public and checkable: Axis's white paper written against the 2025 edition presents a seven-level scale, while its earlier paper written against the 2014 edition presents the familiar four. That is Axis's published material, not our reading of the standard.

The practical consequence for you is simple. If a proposal quotes DORI figures and cites the 2014 edition as current, that proposal was written from an out-of-date reference. It does not automatically make the design wrong — the arithmetic in this article still works — but it tells you something about how carefully the numbers were sourced.

What pixel density does not tell you

Pixel density is necessary. It is not sufficient. A camera can hit 500 px/m and still hand you unusable footage:

• Motion blur. Someone walking briskly through a dim corridor smears across frames if the shutter is slow. The pixels are there; the face is not.
• Backlight. The classic failure is a camera aimed at a glass entrance. At midday the person walking in is a silhouette against a bright doorway, and no amount of resolution recovers a face from a black shape.
• Focus. A lens set once at installation and never re-checked drifts, and nobody notices on a live view they glance at.
• Night. Cameras behave very differently after dark than the daytime demo suggests. The only honest test is to look at footage of a person, at night, in the actual spot that matters.
• Angle. A camera high on a wall looking steeply down at a doorway gets a beautiful view of the top of a hat.

That last one has a fix, and we want to be clear about its status: Guard Nation's own installation standard is to mount face-capture cameras at roughly head height rather than up under the eaves, so the camera sees a face rather than a scalp. That is our practice, based on our results in the field. It is not a building code requirement, not a standard, and not an industry rule. Anyone who tells you a specific mounting height is mandated should be asked to show you where.

What to ask before you buy anything

Four questions, and they work on us as readily as on anyone else:

1. At which specific spots do I need to identify a person, as opposed to just see activity? Usually a shorter list than people expect — often the entrances, the till, and one or two blind corners.
2. What pixel density will this camera deliver at that spot? Not at the sensor. At that spot, at that distance, with that lens. A vendor who designs properly can answer this before a single bracket goes on a wall.
3. Am I paying for resolution that gets thrown away? A 4K camera pointed at a wide yard for general awareness is a legitimate design choice. A 4K camera sold to you as an identification camera for a 30 m scene is not.
4. Can you show me night footage of a person at that spot? Not a brochure. Your site, your lighting, a real person.

The uncomfortable truth is that most sites do not need more megapixels. They need the pixels they already own pointed at fewer square metres. That often means one more camera in the right place rather than a whole new system — which is a cheaper conversation than the one you were bracing for, and a much better one to have before an incident rather than after.