Most security cameras record everything and identify nobody

The short version

• Resolution on its own means nothing. What matters is how many pixels land on the person, and that depends on how wide a scene the camera is asked to cover.
• A 4K camera aimed across a whole parking lot cannot identify a face. Aimed at a doorway, the same camera can.
• After dark most cameras switch to infrared and colour disappears. A navy jacket and a black jacket become the same jacket.
• Mounting height is the cheapest mistake to fix and the most common one. A camera at four metres over a door records hats.
• Thirty days of retention is a habit, not a rule. Pick a number for a reason and write the reason down.

Pixels on target is the only number that matters

There is a published scale for this, so you do not have to argue about it. IEC 62676-4 sets out four levels of pixel density, measured in pixels per metre across the scene. Axis summarises the numbers in its pixel density white paper.

The arithmetic is one division. Take the horizontal pixel count of the camera and divide it by the width of the scene in metres.

A 4K camera is 3840 pixels wide. Point it across a parking lot so the frame spans 30 metres and you get 128 pixels per metre. That is recognition of people you already know, and nothing more. The same camera framed on a 15 metre section of that lot gives 256 pixels per metre, which crosses into identification. Framed on a three metre doorway it gives well over a thousand, which is more than you need.

Run the same sum on a 1080p camera, 1920 pixels wide, across the same 30 metre lot. Sixty four pixels per metre. You will see that a person walked to the truck. You will never learn who.

This is why “we put in 4K” answers nothing. The useful question is how wide each camera is looking, and no brochure tells you that. Your installer should.

The camera cannot see what the light does not show

At night almost every fixed camera drops to monochrome and turns on its infrared illuminators. Two things follow from that, and both matter in an insurance claim.

Colour is gone. Descriptions of clothing, vehicles and hair become guesses in shades of grey.

Infrared light falls off fast with distance. A camera rated for 30 metres of IR is not lighting a face at 30 metres, it is putting a faint glow on a shape. Close subjects near the lens go bright white while anything past the useful range stays black. If a raccoon triggers a bright wash of IR reflecting off a wall two metres away, everything behind it disappears.

Daylight has its own failure. A camera looking at a glass entrance from inside is pointing at the brightest thing in the building. You get a silhouette walking in. Wide dynamic range processing helps and it does not solve it. Turning the camera to look across the doorway rather than straight at it usually does.

The cheapest improvement to most systems is not a camera. It is a light. A wall pack over the loading bay, or a fixture on the gate post, gets the camera out of infrared mode and back into colour. That single change often does more than a resolution upgrade would.

Height and angle decide whether you get a face or a hat

We mount identification cameras high enough to be out of easy reach and low enough that the lens is looking at faces rather than down at scalps. On a door, that is roughly head height plus a bit. The steeper the down angle, the more of the face the brow and the hat brim hide, and past a certain tilt you are photographing shoulders.

The corner-of-the-room mount at ceiling height is fine for an overview camera whose job is to show what happened in the space. It is the wrong mount for the camera whose job is to say who came through the door.

Pick your chokepoints and treat them differently from your general coverage. Every person who enters your building passes through a door frame roughly a metre wide, at walking pace, facing forward. That is the easiest identification shot you will ever get, and most buildings do not have a camera set up to take it.

Frame rate matters less than shutter speed

Fifteen frames a second is enough for almost every scene in a commercial building. Doubling to thirty roughly doubles what you store and buys very little for a person walking.

The setting that actually ruins night footage is shutter speed. In low light the camera holds the shutter open longer to collect enough light, and anything moving smears across the frame. A person at walking pace turns into a blur. A vehicle at yard speed turns into a streak with a suggestion of a plate on it.

You can force a faster shutter, and then the image gets darker and noisier. There is no setting that creates light. The way out is a lamp, which is the same answer as the section above.

Thirty days of retention is a decision, not a default

Retention gets set once, at install, usually to whatever the recorder happened to manage with the drives that came in the box. Then nobody looks again.

Choose it against how long it takes you to notice a problem. A stolen tool gets noticed the next morning, so a week would do. A till discrepancy or a missing pallet gets noticed at month end, and 30 days is then far too tight, because by the time you count you have already lost the start of the month. If your reconciliation is monthly, 45 or 60 days is the honest number.

Two other things pull on it. Insurers and police often come back with questions weeks after the report, so short retention can cost you a claim. And storage scales close to linearly, so doubling the days roughly doubles the disk. That is a real cost but it is a known one, and it is far cheaper than the incident you cannot show anyone.

Pulling the other way: footage of your own staff is personal information, and keeping more of it for longer is not free either. If you employ 25 or more people in Ontario on January 1, you must have a written policy on electronic monitoring of employees in place before March 1 of that year, per the Ontario government’s guide to the Employment Standards Act. Cameras count. The policy has to say how and in what circumstances you monitor, and what you use the information for.

One more thing about retention, and it is the one that catches people. The number configured on the recorder is not the number you have. Cameras drop off, drives fill, motion recording skips events. Actual retention is whatever the oldest file on the disk says it is, and the only way to know is to look.

Seeing that it happened and knowing who did it are two different jobs

Give those jobs to different cameras.

An overview camera covers the space so you can follow what happened and when. It can be wide, it can be modest resolution, and it does not need to see faces. An identification camera sits at a chokepoint, framed tight, mounted at the right height, with light on the subject. It sees almost nothing else and that is the point.

Four cameras placed against those two jobs beat the twelve camera kit spread evenly around the roofline. Even coverage is how you end up with a system that shows everything at 60 pixels per metre.

The audit you can run tonight with your phone

Half an hour, no tools, no technician.

1. Open the live view on your phone. Walk to the spot where an incident would actually happen: the gate, the till, the back door, the trailer.
2. Have someone stand there while you look at the screen. Do not ask “can I see a person”. Ask “could I pick this person out of five strangers”.
3. Repeat after dark, with the yard lights however they normally are. Not with everything switched on for the test.
4. Now check playback rather than live view. Plenty of systems stream one quality to your phone and record another. Look at yesterday, same spot, same question.
5. Read the timestamp on every camera. A clock that drifted two hours makes footage very hard to use and very easy to challenge.
6. Scroll back to the oldest recording on the system. That date, not the setting in the menu, is your retention.
7. Export one clip to a USB stick and time yourself. When police or your insurer ask, they want a file, not a photo of your monitor.

Write down which cameras failed which step. That list is the whole scope of work.

What this means for your building

A contractor’s yard in Courtice is the clearest case. The overview cameras on the corners of the shop do their job: they show a truck came in at 02:14 and left at 02:31. Nobody is identified by them and nobody ever will be. What changes the outcome is one camera on the man door at head height with a light above it, and one camera on the driveway aimed at the pinch point where vehicles slow to turn, framed narrow enough to read a plate.

A dental clinic in Whitby has the opposite problem. Coverage is fine, everything is indoors and lit, and the exposure is retention plus policy. Footage of staff at a reception desk all day is employee monitoring whether or not you think of it that way.

A warehouse loading bay in Ajax usually fails at the bay doors, where a camera mounted up in the structure looks down at the top of a forklift and the driver never appears.

What to do about it

1. Measure the scene width of your three or four most important cameras and do the division. You will learn more in ten minutes than a site visit would tell you.
2. Re-aim before you re-buy. Narrowing the field of view on an existing camera is free and often moves it from observe to identify.
3. Add light where the camera is fighting. Colour footage of a face beats monochrome footage of a shape every time.
4. Nominate your chokepoints and put a purpose-built camera on each one, mounted low enough to see faces.
5. Set retention against how long you take to notice a problem, then confirm the oldest file on disk matches it.
6. If the audit shows gaps you cannot fix by aiming and lighting, that is when a security camera installation is worth quoting, and a CCTV maintenance plan is what stops the same list coming back in two years.