How to reduce false alerts from your security cameras

Everyone with a camera app knows the spiral. It pings for a moth at 2am, then for a passing car, then for rain, and within a fortnight the notifications are muted and the cameras are back to being something you check after it's too late. A muted alert is the same as no alert, so false alerts quietly defeat the point of owning cameras at all.

Why motion alerts fire at nothing

Motion detection compares each frame with the last and fires when enough pixels change. That is the entire analysis. A burglar and a branch produce the same evidence, because the camera knows that something changed and nothing else. So headlights sweep across a wall, a cloud moves the shadows, rain glitters in the infrared, and at night the IR lamp itself attracts insects that cross the lens as huge white streaks. None of that is a malfunction. The camera is answering the question it was asked, and the question was wrong.

Camera-side fixes worth doing anyway

1. 01

   Draw tighter motion zones

   Most cameras let you mark which parts of the frame count. Exclude the road, the sky and anything that waves in the wind, and keep the door, the gate and the drive. This one change removes most passing-traffic alerts on its own.

2. 02

   Drop the sensitivity a notch

   Lower it one step, then live with it for a few nights before lowering it again. Big jumps swap false alerts for missed ones, which is a worse trade than the pings.

3. 03

   Reposition away from the noise

   A camera staring into foliage, or mounted where headlights rake it at every corner, will always cry wolf. A metre of repositioning often beats anything in the settings menu.

4. 04

   Expect insects at night

   The infrared that lights the scene also attracts insects, and a moth a few centimetres from the lens registers as a large bright object. If your false alerts are nocturnal white blurs, that's what they are; our page on why footage looks grainy at night covers what else IR does to the picture.

Person detection helps, but it can't tell you what happened

On-camera person detection is a real improvement. It ignores rain and headlights and fires when the shape in the frame is human, which alone cuts a lot of the nonsense. The trouble is that a person isn't an event. The postman is a person. So is the neighbour taking the bin out, and the pings continue, just with better vocabulary. Knowing a person is present is not the same as knowing something happened worth interrupting your evening for, and that judgement is still yours to make, one notification at a time.

Alert on the event, not the motion

The durable fix is to change what an alert means. Svid's checks have a vision model read footage in plain English and judge a window of recent frames together, so what raises an alert is a described event (someone going down and staying down, an item being concealed, a person inside a zone they shouldn't be in) rather than a pixel change. A branch moving is none of those things, so it doesn't qualify. And you choose which events you want to hear about in the first place, so the till camera watches for concealment while the yard camera watches the restricted zone.

The alerts themselves are built not to become the new noise. One event fires one alert, and the same check firing again within 2 minutes is recorded as a continuation of that event, not a fresh one, so a ten-minute incident is a single entry rather than a barrage. Delivery is the on-screen Alerts feed, native push to your phone, email if you opt in, a JSON webhook into systems you already run, or WhatsApp, and WhatsApp is deliberately capped at one message every 10 seconds per account, so even a genuinely busy site can't flood a phone. The alerts guide covers setting all of that up.

Why does my camera keep alerting when nothing's there?

Something is there, just nothing that matters: rain lit by the infrared, a headlight sweep, an insect near the lens, a shadow moving with the clouds. Motion detection can't rank meaning, so it reports them all. Tighter zones and lower sensitivity cut the volume; alerting on described events removes the category.

What motion sensitivity should I set?

There's no universal number, because it depends on the lens, the scene and the mounting. Change it one step at a time and judge each step over a few nights, and draw tighter zones before touching sensitivity at all, since the zones usually do more.

Do cameras with person detection stop false alerts?

They stop a lot of them, and they're worth having. What they can't do is tell you whether anything happened; a delivery driver and an intruder are both a person. You'll still be triaging pings by eye, just fewer of them.

Won't AI alerts flood me the same way?

The structure is different. An alert is raised on a described event you chose to watch for, one event fires one alert, and a re-fire within 2 minutes is recorded as a continuation rather than a new alert. WhatsApp delivery is capped at one message every 10 seconds per account on top of that.