Walk into almost any commercial building in the Lower Mainland, look up, and you will find the cameras in the same place: the high corner of the room, up near the ceiling, angled down across the floor.
There are good reasons for that. It is the cheapest place to mount a camera, the easiest place to reach with cable, and the hardest place for anyone else to reach without a ladder. From up there one camera sees a lot of floor. Every incentive in the installation points at the corner.
It is also, geometrically, the position least likely to give you a face.
That is the trade this article is about. It is not an argument that corner cameras are wrong. It is an argument that a site built entirely out of corner cameras has quietly chosen one capability and given up another, usually without anyone saying so out loud.
What the corner is actually good at
A corner camera does one job extremely well: it tells you the story of a space.
It shows you how many people were in the room, where they went, what they touched, how long they stayed, and whether anyone else was with them. When you are reconstructing an incident, the sequence of events is often the part that matters most — the identity question may already be settled, or may not be answerable by any camera.
The corner is also where the camera survives: out of arm's reach, out of the way of doors and carts and forklifts, and not in anyone's personal space. Cameras that people can touch get touched.
So the corner earns its place. The problem is what happens when it is the only thing on the drawing.
Why the corner loses the face
Two separate effects work against you, and they compound.
The first is angle. A camera high on a wall is looking down at a steep angle. A person walking underneath it presents the crown of their head, the brim of a cap, the shoulders of a jacket — and the face, which is a roughly vertical surface, is turned away from the lens. The steeper the look-down, the more of the face is hidden by the person's own forehead, hat and hood. Nothing downstream fixes this. The information never reached the sensor.
Nobody publishes a face-capture angle limit we are willing to quote at you, and we are not going to invent one. But you can see the shape of the relationship in a vendor's published figures for a much simpler machine-vision task. For its own licence-plate product, Axis states that "the camera's mounting angle should not be larger than 30 degrees in any direction" and that "the image of the licence plate should not tilt more than 5 degrees horizontally," with a rule of thumb that capture distance is roughly twice the mounting height. A plate is a flat, high-contrast rectangle designed to be read — and even that has a published angle budget, and a mounting height that has to be traded against distance.
A human face is a curved, low-contrast, moving, self-shadowing target that can be pointed anywhere. It is harder than a plate, not easier. The transferable part is the direction of the relationship — the steeper the camera looks down, the less of the front of a person it can see. Those specific figures are Axis's numbers for Axis's plate-reading analytic, not a rule for faces.
The second is distance. A camera that covers a lot of floor is, by definition, spreading its pixels across a lot of floor. That is the pixel density problem, and it is the other half of this one — a camera can be at a perfect angle and still hand you a smear, because the person occupied too few pixels. We work through the arithmetic in Pixel density: what "enough resolution" actually means, and it is worth reading alongside this, because the two failures are usually diagnosed as one vague complaint: "the footage is useless."
The corner tends to maximise both problems at once. It is the furthest point from most of the room and the steepest angle onto anyone in it.
Choke points: the concept that makes this manageable
The instinct, once you see the problem, is to want faces everywhere. You cannot have that, and you do not need it.
What you need is choke points.
A choke point is any place where movement through your site is compressed — where people cannot spread out, must pass in a predictable direction, and are moving at a predictable speed. Every building has a handful of them, and they are usually obvious once you go looking:
- the front entrance and any staff or delivery entrance
- the till, counter, or dispensing window
- a corridor between two functional areas
- the bottom of a stairwell, or the elevator lobby
- a gate, man-door, or turnstile in a fenced yard
- the point where a loading bay narrows to a doorway
Choke points are valuable for four reasons at once, and those four are what make a single camera there worth several elsewhere.
They narrow the scene. A doorway is a couple of metres wide. A yard is tens of metres wide. The same camera pointed at the narrower scene puts far more pixels on whoever passes through it.
They fix direction of travel. In an open space a person can face any way. Through a door, almost everyone is facing the same direction — forwards, into the building. A camera positioned to look back along the path of travel meets them face-on rather than from above and behind.
They fix the distance. You know within a metre or so where the subject will be when they cross the plane of the door. That is what makes the arithmetic answerable in advance, instead of a hope.
They fix the lighting. One spot, one lighting condition, testable. A camera covering a whole yard has to cope with every lighting condition in that yard simultaneously.
The reason a camera at roughly head height on a choke point outperforms a corner camera for identification is that it improves all four variables in one move — closer, narrower, face-on, and in known light.
Both kinds. The mistake is having only one.
Here is the part that gets lost when this argument is made badly.
A face-height camera on a door is close to useless for telling you what happened in the room. Its field of view is small on purpose. It sees a person for a second or two, front-on, and then they are gone. It cannot show you where they went, what they did, or who they were with. If you build a site entirely out of identification cameras at choke points, you will be able to say who came in and completely unable to say what they did.
The corner camera answers that second question, and no face camera replaces it.
So the correct design has both, deliberately, with each camera assigned one job:
- Coverage cameras in the corners, wide, high, tamper-resistant, telling the story of the space.
- Identification cameras on the choke points, closer to head height, looking back along the path of travel, narrow field of view, covering one thing well.
The house rule we use — and we want to be precise that this is Guard Nation's own installation practice, based on our results in the field, not a code requirement or an industry standard — is that a camera whose job is to capture a face gets mounted at roughly head height and aimed along the path of travel, rather than under the eaves looking down. Anybody who tells you a specific mounting height is mandated should be asked to show you where it is written.
The failure mode we see most often on existing sites is not a bad camera. It is eight corner cameras and zero identification cameras, on a site whose owner believes they have eight of each.
Thermal cameras make the same trade unusually honest. Axis says of its thermal range that "unlike conventional cameras, this technology only allows for detection," and that while it can determine a person's shape and size, "it doesn't allow for clear identification." A corner camera is not thermal, but the category it usually falls into is the same one: detection and context, not identification.
The honest costs of the face-height camera
We would be selling you something if we left it there. Cameras at head height have real disadvantages, and you should hear them before you buy any.
They are reachable. A camera someone can touch is a camera someone can cover, turn, tape over, or damage. This is the single strongest argument for the corner, and it is not a small one.
They can be blocked. A queue, an open door, a stacked pallet, a delivery trolley — anything that sits between the camera and the choke point takes the camera out of service, and nobody notices until they need the footage.
They are visible, and that cuts both ways. A camera at eye level is obvious to everyone who walks past. Some owners want that; others find it changes the feel of a reception area in a way they do not like.
There is not always a surface. A glass vestibule, a curtain wall, a heritage façade, a leased space where you cannot core the wall — sometimes the position that would work is one you cannot mount to.
Backlight can get worse, not better. Aiming along the path of travel at an entrance often means aiming at the doorway, and at midday the person walking in is a silhouette against a bright opening. Position solves the angle problem and can make the exposure problem harder. The only test that settles it is footage of a real person, at that spot, in the worst light that spot gets.
The privacy half, which is not optional
A camera that reliably captures identifiable faces is collecting more personal information than a camera that captures the top of a hat. In British Columbia that is a legal question and not only a technical one.
Under the Personal Information Protection Act, a private-sector organization has notification duties around the collection, use and disclosure of personal information at sections 13, 16 and 19, security obligations at Section 34, and retention and destruction obligations at Section 35. The practical consequence for camera placement is that the identification camera is the one most likely to make signage, retention limits and access controls matter, because it is the one actually producing identifiable images.
Separately, the Privacy Act makes surveillance a possible route to a privacy violation, whether or not there is any trespass, at Section 1(4) — and sets the standard at Section 1(2) as the privacy that is "reasonable in the circumstances, giving due regard to the lawful interests of others." A camera positioned to capture faces at your own entrance sits comfortably inside that. A camera positioned at head height covering a neighbouring property, a residential window, or a washroom approach does not, and no amount of legitimate security purpose upstream repairs it.
We are describing the shape of the obligation, not giving legal advice. If a placement is close to a line, the placement is the cheapest thing to change.
What to ask before the drawings are final
Five questions. They work on any proposal, including ours.
- Which cameras on this drawing are identification cameras, and which are coverage cameras? If the answer is "they all do both," the design has not been thought about. Ask for the list, camera by camera.
- Where are my choke points, and is there a camera on each? You can answer the first half yourself, on foot, in ten minutes.
- What is this camera's job in one sentence? A camera that cannot be described this way is usually covering nothing in particular at a distance that guarantees it.
- What does the face-capture camera see when the door is propped open, or when there is a queue? The everyday condition, not the empty-building one.
- Can you show me footage of a person at this spot, at night, from this position? Not a brochure. Your site.
The reason to have this conversation before an incident rather than after is that afterwards, the only available remedy is regret. Beforehand, the fix is frequently one more camera in the right place rather than a new system — and that is the cheaper conversation of the two.
