How far can a security camera detect, recognise and identify a person?

There is no single detection range for a security camera, because "can it see that far" is really four different questions with four different answers. The industry standard that separates them is DORI — Detect, Observe, Recognise, Identify — defined in IEC 62676-4 (published in Europe as EN 62676-4), the international standard for video surveillance system application guidelines. DORI states the minimum pixel density needed on the subject for each task, and everything else on this page follows from it.

What each camera gives you, in feet

Take the camera's horizontal pixel count and divide by how wide the view is at the point you care about. The table below runs that arithmetic backwards: for each resolution, the widest slice of scene that still meets each DORI threshold.

Read it as a coverage budget. A single 4K camera can detect movement across a 500-foot frontage, or identify a face across a 50-foot gateway — but not both, and not at once. This is why sites are usually covered by a wide camera that catches the event and a narrow one aimed at the pinch point people have to pass through.

How many megapixels do I need to identify someone at 100 feet?

Fewer than you would think, if you pick the right lens — and no number of megapixels will do it with the wrong one. Identification needs roughly 76 pixels per foot of scene. At 100 feet away, an ordinary 1080p camera fitted with a long lens — around 25 mm on a common 1/2.8-inch sensor, giving a field of view near 13 degrees — covers a slice about 22 feet wide and clears the identify threshold comfortably. The same 1080p sensor behind a wide 2.8 mm lens covers well over 100 feet of scene at that distance and gives you barely detection.

So the honest answer to "how many megapixels" is: megapixels buy you width at a given quality, not distance. Going from 1080p to 4K does not let you see twice as far — it lets you cover twice as wide a scene at the same pixel density, or keep the same scene at double the density. Both are useful. Neither is a zoom lens.

Can cameras read licence plates entering a lot?

Yes, but only from a camera set up to do that one job — a general overview camera will not do it, whatever its resolution. Plate capture is closer to a measurement instrument than to surveillance, and it has four hard requirements.

• A dedicated, tightly-framed camera. The plate needs to fill a meaningful part of the frame, which means a narrow field of view aimed at one lane at one point — typically a gate, a barrier or an entry throat.
• A shallow angle. Beyond roughly 30 degrees off the plate's face, horizontally or vertically, character shapes distort and read rates fall away fast.
• A fast shutter. Plates on moving vehicles need a much faster shutter than a general scene, which makes the image dark. That is expected and it is why plate cameras look wrong when you view them as ordinary footage.
• Infrared, usually. Licence plates are retro-reflective, so an IR illuminator makes the plate glow against a dark background and produces the cleanest reads at night — often better than daylight.

Plan for a plate camera per lane, per direction. And be clear internally about what you want it for: a record of who entered, which is achievable, is a different system from real-time matching against a list, which adds an integration and a data-protection conversation.

Can security cameras detect a gun or a weapon?

Not reliably, and we do not offer it. It is worth being direct about this because it is marketed heavily. Detecting a weapon in video means detecting a small, frequently dark object that is often partly occluded by a hand or a body, at whatever range and pixel density the camera happens to give — and the DORI table above shows how quickly that density disappears. Systems that demonstrate well do so with a large firearm held clear of the body, unobstructed, in good light, close to the camera. Concealed weapons are outside what video can see at all.

The failure modes cut both ways and both are expensive. A missed weapon is the scenario the system was bought for. A false positive sends an armed police response to a contractor holding a cordless drill. Where sites want early warning of a serious threat, the reliable signals are behavioural and contextual — someone entering a restricted area outside hours, a vehicle waiting at a boundary, a person moving against the normal flow — and a trained operator watching a live view and making the call.

Do AI cameras work in rain, fog and dust?

They keep working, but their effective range shortens, and the reason is optical rather than algorithmic: detection quality can only be as good as the image. Airborne water and dust scatter light, which reduces contrast between a subject and the background — the exact signal a classifier depends on. Assume a materially shorter usable range in bad weather than the specification sheet implies, and design coverage so the events you care about happen close to a camera rather than at the edge of one.

• Rain at night is the hardest case. Infrared illumination bounces off droplets close to the lens, producing bright streaks that drift across frame and generate false alerts. Separating the illuminator from the camera body, or lighting the scene with white light instead, fixes most of it.
• Fog kills contrast before it kills resolution. The subject is still in frame at full resolution but has nearly the same brightness as the air around it. Thermal imaging degrades far more gracefully here, because it is reading heat rather than reflected light.
• Dust and pollen coat the lens and the dome. A gradual haze that nobody notices until footage is needed. On construction and agricultural sites this is the single most common cause of a system quietly getting worse.
• Wind is an underrated failure. Vegetation, banners and loose fencing move continuously, and a pole that flexes in gusts makes the whole scene move. Rigid mounting matters as much as the camera.

Thermal, infrared or low-light: which do I need at night?

They solve different problems and are frequently combined. Choose by what has to happen after the detection.

The common pairing on an outdoor site is thermal to detect at range and an IR or low-light camera at the approaches to give a person something to act on. Thermal alone tells you someone is there; it will not tell an operator, or later a police officer, who. If you only buy one, buy for the identification job and light the scene properly — but expect a shorter reliable range than a thermal unit would give you.