Stealth Games Explain Where Your Cameras Stop Seeing

A homeowner sent me a clip from her doorbell camera last autumn. A man walks up her path, takes a parcel and leaves. The camera caught all of it, in colour, at night. She could not identify him at all. His face filled maybe a dozen pixels, and when he bent down for the box, the top of his head was the clearest thing in frame.

When I installed physical security systems professionally, I spent a lot of time explaining to clients that "the camera saw it" and "the camera can tell you who it was" are different promises. These days I explain it with stealth games, because a surprising number of people have played one, and the genre is built entirely around the gap between what a watcher can detect and what it can actually work out. Here is how that gap looks in games, and how to measure it on your own front door.

What stealth games teach about being watched

Thief: The Dark Project in 1998 gave players a "light gem" that showed how visible they were. Metal Gear Solid drew guards' vision as cones on a radar. Splinter Cell added a visibility meter and cameras on walls that swept back and forth. Across all of them, players learned a few rules very quickly.

Watchers see in a cone, not a circle. Guards rarely look straight up or straight down. Cameras have a sweep pattern with a moment of blindness at each end. Darkness and backlighting both help you hide. And most importantly, the first stage of being seen is almost never "identified." It is "something moved," followed by a guard walking over to investigate.

Home cameras follow almost every one of those rules. The difference is that the people who designed the games wanted you to find the gaps, and the people who sold you the camera did not.

Detecting is not identifying

Professional installers use a standard called IEC 62676-4, which sets out how many pixels per metre of scene a camera needs for different tasks. The industry shorthand is DORI:

The distances come from simple geometry. A 4 megapixel camera is roughly 2,560 pixels wide. At a 110 degree horizontal field of view, the scene it covers is about 2.86 metres wide for every metre of distance. Divide one by the other and you get roughly 896 pixels per metre at one metre away, falling as you go further. Doorbells with 160 degree lenses spread the same pixels over a much wider scene, so their identify range shrinks to well under two metres.

That was the problem with her clip. The man never came closer than about four metres to her doorbell, and he looked down most of the time. In game terms, the camera had spotted him and could have "observed" him, but it never got close enough to identify.

Where the blind spots really sit

Stealth players instinctively hug walls directly under a camera, because the lens cannot tilt that far down. Real cameras mounted high have the same flaw.

Under the camera

A lot of people mount cameras at 3 metres or more to keep them out of reach. Mount one at 3 metres with a 50 degree vertical view angled slightly down, and there is a wedge on the ground near the wall, often a metre or more deep, the camera simply cannot see. Worse, anyone who walks under it is filmed from directly above. You get a lovely view of a baseball cap.

For a front door, I aim for about 2.2 to 2.5 metres high, angled so a person's face is in frame from roughly 1.5 to 4 metres out.

Head on approach

Motion detection on most battery cameras uses a passive infrared sensor, which reacts to a warm body crossing between its detection zones. That is the light gem principle in reverse: moving across the field is loud, moving straight towards the sensor is quiet. A person walking straight up a path at the camera can get within a couple of metres before triggering it. By the time recording starts, the best identify window may be over. I go into the sensing side in more depth in what a motion sensor alert actually tells you.

Backlight and the evening shift

In Splinter Cell, a bright lamp behind you makes you a silhouette. The same happens to cameras facing a street with low winter sun, or a porch with a bright lamp next to the lens. Faces go dark against a bright background. If your camera faces west, check footage from late afternoon in winter, not just midday in summer.

The advice I would push back on

The usual advice is to buy a camera with a wider field of view so "nothing gets past." I think that is mostly backwards for a front door. A wide view is good for detecting. It is bad for identifying, because every pixel covers more of the world. Many of the clients I worked with would have been better off with a narrower camera pointed at the exact spot a visitor must stand, the doorstep, plus a cheap wide one covering the drive. Two cameras, two jobs.

A camera that sees the whole garden often sees nobody's face.

Placement matters more than the spec sheet, and I wrote a longer piece about where to point a doorbell camera with diagrams of common front door layouts.

Walk your own patrol route

This is the stealth game exercise, done in reverse. Get someone in the house to watch the live view while you walk up to your front door the way a stranger would: down the path, straight at the door, then bending down as if picking up a parcel. Walk it again in a hat. Then do it at dusk. Note at what distance the camera triggers, at what distance your face becomes recognisable, and whether bending down hides it. Most people find at least one gap in under ten minutes. Fixing it is usually a matter of lowering the camera, changing its angle, or adding a second camera aimed at the doorstep. If the walk turns up lots of false triggers instead, handling false alarms from a smart security system is the next thing to read.