Motion Detection and Detection Zones

What is a detection zone, and why does it matter more than sensitivity?

A detection zone is a drawn boundary inside the camera's full field of view that limits motion alerts to that area only. Sensitivity, by contrast, is a single number applied across the whole frame, zone or not, that decides how much pixel change counts as motion.

A zone shaped correctly excludes the street, the neighbour's yard and any moving foliage before sensitivity ever enters the picture, which is why fixing zone shape resolves far more false alerts than any sensitivity adjustment does.

This split also explains why two cameras with identical hardware can behave completely differently once installed. The lens, sensor and firmware are fixed at the factory, but the zone is drawn by whoever sets up the camera, which means most of the false alert problem is actually a configuration problem rather than a hardware one.

Why do most false alerts come from zone shape rather than the camera?

A zone that extends past the property line captures every car and pedestrian on the street. A zone that crosses a tree line captures every gust of wind. Neither problem is a camera defect, both are a zone drawn too large or too far, and both disappear once the zone is redrawn to the actual approach path a person would use.

The same logic that makes an oversized zone a problem also makes a zone drawn to the wrong shape a problem even at the correct size. A rectangular zone stretched over a diagonal driveway captures a triangular slice of unrelated ground on either side, while a custom polygon traced to the actual edges of the driveway removes that ground from consideration entirely without touching sensitivity at all.

How many pixels does a camera need to reliably see a person move?

The EN 62676-4 detect threshold, 25 pixels per metre of scene width, sets the practical floor. A person roughly 1.7 m (about 5 ft 7 in) tall occupies about 43 pixels of vertical height in the frame at that threshold, which is the smallest silhouette most motion and AI person detection engines can work with reliably.

Below that pixel height, at a distance beyond the camera and lens combination's detect distance, a person and a stray cat can look equally ambiguous to the algorithm, and no sensitivity setting fixes a target that is simply too small in the frame.

That 43 pixel figure is not a hard cutoff so much as a floor below which reliability drops sharply. A target at 50 or 60 pixels of height still benefits from a wider margin against noise, motion blur and compression artifacts than a target sitting right at the 43 pixel line, which is one more reason a zone drawn well inside a camera's detect range outperforms one drawn at the very edge of it.

How does distance change a target's pixel footprint?

Pixels per metre falls off with distance for the same reason DORI distance is defined the way it is: the same lens spreads a fixed number of horizontal pixels across a wider slice of scene the farther away the target stands. A person at the edge of a zone that just barely clears the detect threshold has a razor thin margin, and a small increase in distance, or a slightly wider lens, can push them below reliable detection.

This is also why a zone drawn to the exact edge of a property, rather than to the actual walking path a visitor or intruder would use, wastes the camera's best pixel density on ground nobody actually crosses.

The same math explains why a person walking directly toward the camera changes size dramatically over just a few steps near the edge of a zone, while a person walking across the frame at a constant distance keeps a roughly constant pixel height throughout. A zone drawn along a walkway that runs toward the camera should account for the far end of that path having noticeably fewer pixels than the near end.

Should sensitivity or zone shape be adjusted first?

Zone shape first, sensitivity second. Raising sensitivity to catch a distant or marginal target usually just adds false alerts from wind, shadows and headlights across the rest of the frame, since sensitivity is not aware of where in the frame the motion happened.

Shrinking the zone to the specific approach path, a walkway, a driveway lane, a gate, keeps the camera's attention on the ground that actually matters and lets the default sensitivity setting do its job without picking up everything else in the field of view.

Time of day schedules are a second, complementary tool: a driveway camera can run a wider zone and higher sensitivity during a work day when nobody is expected home, then automatically switch to a tighter zone and standard sensitivity overnight, layering a schedule on top of the zone rather than trying to make one static setting do both jobs.

A camera mounted too high compounds the same problem from a different angle. Tilting a camera downward to see a walkway directly below it shortens the effective distance to that ground and pushes anything beyond a few feet toward the edge of the field of view, where zone accuracy and pixel density both suffer at once.

Does AI person and vehicle detection change any of this?

AI classifiers still need the same pixel floor as basic motion detection to tell a person from a vehicle or an animal, they are simply better at making that call once enough pixels exist. A target below roughly the 43 pixel detect height is as hard for an AI classifier as it is for simple frame differencing.

AI detection genuinely reduces false alerts caused by wind, insects near the lens and small animals within its reliable pixel range, which is why it shows up on cameras built around that feature, but it does not rescue a target that is simply too far away for the lens and resolution combination in use.

AI detection models are typically trained on people and vehicles photographed at a range of distances and angles, but every one of those training images still had to contain enough resolved pixels to be labeled accurately in the first place. A model cannot learn to recognize a pattern that was never resolvable in its training data, which is the underlying reason the pixel floor applies to AI detection just as much as it applies to simple motion detection.