How to reduce CCTV false alarms

Why false alarms happen

A basic motion rule compares one frame with the next and raises an alert when enough pixels change. It has no idea what changed. A raindrop, a moth, a sweep of headlights and a person climbing a fence all look the same to it: a change in pixels.

That is why the same few causes turn up on almost every site. Pick the one you see most often to find out why it triggers and what to try first.

Quick fixes: angle, IR reflection, cobwebs, zones and sensitivity

Before changing any software, fix what the camera physically sees. These six steps are in the order most installers work through them, cheapest first.

1. 1

   Clean and check every lens

   Clean the lens and dome inside and out, remove cobwebs and check that each housing is sealed. Dirt and webs are the cheapest cause to fix.

2. 2

   Fix the camera angle

   Point cameras so nothing sits close to the lens, keep roads and reflective surfaces out of view, and tilt down so the sky takes up less of the frame.

3. 3

   Draw tighter detection zones

   Only detect where something matters, such as a gate, a fence line or a doorway. Mask out roads, trees and roof lines.

4. 4

   Set sensitivity and schedules per camera

   Tune each camera separately rather than using one setting for the whole site, and only arm zones in the hours you need them.

5. 5

   Move from motion to classification

   Use detection that recognises people, vehicles, smoke and flame, instead of rules that react to any change in pixels.

6. 6

   Use feedback so the system learns your site

   Mark false detections with the feedback button so the AI fitted to your site is retrained on what is normal for each camera.

Two of these steps deserve more detail. Infrared reflection is the one most often missed: a camera mounted tight under a soffit or beside a wall throws its own infrared light back into the lens, which turns the night image grey and noisy. Moving the camera a short distance away from the surface often clears it completely. Cobwebs are the most common cause at night, because the infrared LEDs attract insects. Cleaning on a schedule, rather than waiting for a run of alerts, keeps them under control.

Zones and sensitivity work best camera by camera. A camera on an exposed corner in the rain needs different settings from one inside a loading bay, and a single site-wide setting usually means one camera is too sensitive and another misses things.

Why motion detection hits a ceiling

Once the lenses are clean and the zones are tight, motion detection still has a hard limit: it cannot tell what moved. Turn sensitivity down far enough to stop rain and spiders, and it also stops reacting to a person at the far end of the yard.

That trade-off is why many sites end up switching motion alerts off at night, or why a monitoring centre sees the same camera trigger again and again. The settings can move the balance, but they cannot remove it. Use the toggle to compare a typical night of motion alerts with the same night using classification.

How AI classification differs

AI detection looks at what is in the image rather than how many pixels changed. It is trained to recognise people, vehicles, smoke and flame, so a moving shadow or a spider on the lens is not a person, however much of the frame it covers.

That changes what an alert means. Instead of "something moved on camera 5", the alert says what the camera saw and comes with the clip, so whoever receives it can check it in seconds. You still draw zones and set schedules, but they decide where and when the AI watches rather than doing all the filtering. Read more in AI intrusion detection and video AI vs thermal vs on-camera analytics.

Classification is not perfect. Some objects look like others at a distance or at night, and a camera that cannot see clearly limits any system. That is why the next step matters.

Why per-site, self-learning AI does better

A generic model has never seen your site. It does not know that the vent on the north wall puts out steam every morning, or that a flag by the gate moves all day. A self-learning AI fitted to your site learns what is normal for each camera.

Every Woodcock AI detection has a feedback button. When someone marks a detection as a false alarm, that feedback goes into retraining for your site, so the same kind of mistake becomes less likely on that camera. Over time the alerts that remain are the ones worth looking at. See how it works for where video is processed: on an edge server on site or in the Woodcock AI Cloud.

What it means for monitoring centres

For a monitoring centre, every false alarm is an operator looking at a camera for nothing. Fewer alerts, each with a description and a clip, are faster to verify and easier to trust.

Alerts to a monitoring centre (ARC) and phone-call alerts are available as optional add-ons. If your site already works with a monitoring centre, agree with them what should count as an alarm on each camera, and review the cameras that trigger most often together. For smoke and fire, the same approach applies: an alert with the clip is quicker to check. More in the video smoke detection guide.

False alarm checklist

Tick off each item as you go. Your progress is saved in this browser only.

• Every lens and dome cleaned and checked for webs
• No wall, soffit or plant within a metre of the lens
• Roads, trees and reflective surfaces masked out
• Zones drawn only where an alert matters
• Sensitivity and schedules set camera by camera
• Alerts include a clip so they can be checked quickly
• False detections marked with feedback
• Monitoring centre, if used, agreed on what counts as an alarm

0 of 8 done