Most false alerts come from one thing: the camera is detecting changed pixels, not people. Rain, a shadow moving across a wall, headlights sweeping a fence, a spider on the lens at 3 AM — all of those change pixels, and a basic motion detector cannot tell them apart from a person walking up your driveway.
Here is what causes each of the common ones, and what actually fixes them.
The real problem with false alerts
They are not just annoying. They are how a security system stops working.
After a couple of weeks of alerts about nothing, you stop looking at them. Then you mute the app. Then the one alert that mattered arrives and it looks exactly like the four hundred that did not. A system you have learned to ignore provides no security at all — it just produces notifications.
So this is worth fixing properly rather than living with.
Cause 1: pixel motion instead of object detection
The oldest and most common method. The camera compares each frame to the last one, and if enough pixels changed, that is "motion".
It has no idea what a person is. It triggers on:
- Rain and snow, especially lit by an infrared LED at night
- Shadows moving as the sun tracks across the day
- Headlights and reflections
- Trees and long grass in wind
- Insects and spiders on or near the lens
- Your own infrared illuminator catching a moth
The fix is object detection — a model that classifies what it is seeing as a person, a vehicle or an animal, and only alerts on the categories you care about. Frigate, which runs on the NoBreak Brain Box, does exactly this in real time and locally: a moth is not a person, so it is not an alert.
If your camera only offers a "motion sensitivity" slider and no object types, it is doing pixel comparison, and no amount of tuning that slider will make it understand what a person is.
Cause 2: the camera is pointed badly
A surprising share of false alerts are an aiming problem rather than a software one.
Common mistakes:
- A road or sidewalk in frame. Every car and every passer-by is a legitimate detection of something you do not care about.
- Aimed at the sky or into the sun. Exposure hunts constantly; the whole frame changes brightness; everything reads as motion.
- Too wide. A camera covering the entire yard captures every leaf. Two cameras covering the two ways in are worth more than one covering everything.
- A tree branch or flag inside the frame. Guaranteed motion on every windy day.
- Mounted too high. A camera at roof height gives you the top of a head. Two to three metres is the usual sweet spot — high enough to be out of reach, low enough to record a face.
Fixing aim is free and it is usually the single biggest improvement available.
Cause 3: no zones configured
Even with good object detection, you probably do not want to know about every person your camera can see — the neighbour on their own driveway, pedestrians on the street.
Zones restrict alerts to areas that matter: the path to your door, the driveway, the side gate. Person detected on the public sidewalk, no alert. Same person on your walkway, alert.
This is one of the highest-value ten minutes of configuration in any camera system, and most installations never do it.
Cause 4: night-time infrared
Cameras behave very differently after dark, and most false-alert complaints are night complaints.
- Insects are attracted to the IR illuminator. A moth centimetres from the lens fills the frame and looks enormous.
- Rain and snow catch the IR and light up as bright streaks — a blizzard of fast-moving bright objects.
- IR reflects off nearby surfaces — a wall, a soffit, a window — washing out the image and making the exposure hunt.
Practical fixes: move the camera away from the wall it is mounted near, use external illumination rather than the on-board IR ring where you can, and put the camera under an overhang so precipitation is out of the near field.
Cause 5: context the camera does not have
This is the interesting one, and it is what separates a camera from a security system.
A person on your porch at 2 PM on a Tuesday is a delivery. The same person at 3 AM is not. A car in the driveway when you are home is you. The same car when the system is armed and you are two hundred kilometres away is something else.
Object detection alone cannot make that distinction — it sees "person" both times. What is needed is a second layer that reasons about time of day, armed status, and whether this fits the household's normal pattern.
That is how NoBreak is built: Frigate does fast object classification, and a separate local AI model reasons about the context around it. Critical classification runs in under 2 seconds; contextual analysis runs in under 15. And because both run on the Brain Box in your utility room, none of that reasoning depends on an internet connection or sends footage anywhere.
Low-confidence events get filtered out rather than pushed to your phone, and the system improves as homeowners mark alerts that were wrong.
What you can fix today, in order
- Re-aim the cameras. Get roads, sidewalks, sky and branches out of frame. Free, and usually the biggest win.
- Mount at two to three metres, angled down at the approach.
- Turn on object detection and alert only on person and vehicle, if your camera offers it.
- Draw zones. Ten minutes, large effect.
- Deal with the night. Move the camera off the wall, add an overhang, use external light instead of the IR ring.
- Set different rules for armed and disarmed. You do not need to know about your own family arriving home.
If you have done all six and the alerts are still constant, the camera is doing pixel comparison and no configuration will fix that. That is a hardware limitation, not a settings problem.
The standard we hold ourselves to
An alert should mean something happened that you would want to know about. If it does not, the system is training you to ignore it, and a system you ignore is not protecting anything.
We would rather tune a system until its alerts are trustworthy than ship one that notifies constantly and calls that vigilance.
If your current cameras are burying you in notifications, that is a reasonable thing to bring to a free 15-minute call — including if the answer turns out to be "re-aim the two you have." Toronto and the GTA.
Frequently asked questions
Why does my security camera detect motion when nothing is there? Because basic motion detection compares pixels between frames rather than recognising objects. Rain, shadows, headlights, wind in branches and insects on the lens all change pixels, so they all read as motion to a camera that has no concept of a person.
How do I stop false motion alerts? Re-aim the camera so roads, sky and branches are out of frame, mount it two to three metres up angled at the approach, enable object detection for person and vehicle only, and draw zones so alerts fire on your property rather than the sidewalk.
Why do cameras alert on rain and shadows at night? The infrared illuminator lights up raindrops and snowflakes as bright fast-moving streaks, and attracts insects that fill the frame at close range. Moving the camera off the wall it is mounted near and putting it under an overhang both help substantially.
What is the difference between motion detection and object detection? Motion detection notices that pixels changed. Object detection classifies what changed as a person, vehicle or animal. Only the second can ignore a moth while alerting on someone in your driveway.
Can AI tell the difference between a person and my pet? Object detection distinguishes people, vehicles and animals directly. NoBreak runs Frigate for that classification plus a separate local AI model that reasons about context — time of day, armed status and household patterns — so a familiar pattern does not become an alert.
Do the alerts get better over time? Yes. Low-confidence events are filtered out automatically, and the system improves as homeowners mark alerts that were wrong. Zone and schedule tuning during and after installation is part of the job rather than something left to you.