Alarm fatigue — why motion alerts get ignored, and how to fix it

Ask anyone who has worked a night shift in a monitoring centre what the hardest part is, and the answer is rarely the incidents. It's everything that isn't one: the spider on the lens, the tree in the wind, the headlights across the yard, the fox that walks the same fence line at 2am every night.

That's alarm fatigue. When most alerts are false, people learn, reasonably, to treat them all as false. And that's how the real one gets missed.

Where false alarms come from

Most camera alerts are still triggered by motion detection: pixels changing between frames. Motion detection can't tell what moved. So it fires on:

• Weather: rain, snow, fog, wind moving foliage and flags
• Light: headlights, clouds passing the sun, security lights switching on, the camera's own infrared switching
• Animals: foxes, cats, birds, insects on or near the lens
• Scene changes: shadows moving through the day, reflections in puddles and glass

Tuning sensitivity down trades false alarms for missed ones. Masking areas out creates blind spots. Neither fixes the underlying problem: the system doesn't know what it's looking at.

What false alarms cost

• Operator attention. Every alert takes time to check. A monitoring centre with thousands of cameras can spend most of its capacity dismissing nothing.
• Response. A site that alarms every night gets a slower response when it matters, from operators, keyholders and the police alike.
• Money. Guard call-outs, keyholder visits, and in some cases a lower police response level for sites with a history of false activations.
• People. Night after night of pointless alerts wears down good operators, and they leave.

What actually brings them down

Object detection instead of motion detection. A model that recognises people and vehicles ignores everything else. Rain isn't a person. A fox isn't a vehicle. This single change removes most false alarms on most sites.

Rules that match what you care about. A person in the yard at 3pm is a delivery. At 3am it's an intruder. Detection zones, schedules, line crossings ("entered from the road side"), and dwell time ("stayed near the fence for more than 30 seconds") turn "a person was seen" into "something that matters happened".

A minimum size and confidence. Ignore detections too small to be a person at that distance, or too uncertain to act on, and tune them per camera rather than site-wide.

Learning from dismissals. When an operator dismisses an alert as false, that should feed back: which camera, what time, what triggered it. A handful of cameras usually cause most of the noise, and fixing those few changes the whole site.

Measure it

Before changing anything, count a week of alerts per camera and how many were real. After, count again. "False alarms fell from 400 a night to 12" is the number that convinces a client, an insurer or a monitoring centre. It's also how you find the three cameras still generating most of the noise.

Alarm fatigue isn't an operator problem. It's a detection problem, and it's fixable without replacing a single camera.