Definition
A false alarm is a report of an event that did not actually occur. In safety systems it is a more serious problem than an occasional miss, because it erodes operators' trust in every subsequent report.
Why it matters more than raw accuracy
A system raising a hundred reports a day, five of them real, will be ignored within a week. The same installation with five reports a day, four of them correct, will be taken seriously.
Tuning therefore focuses on removing irrelevant reports rather than raising sensitivity. The goal is an alert somebody will actually act on.
What reduces them in practice
Context helps most: zone rules, schedules, exceptions for maintenance work and a minimum event duration. Only after that does the model's confidence threshold matter.
The second factor is image quality. Correcting a camera's mounting or lighting an area removes more errors than changing the algorithm, and costs less.
Frequently asked questions
What false alarm rate is acceptable?
It depends on the scenario and the number of real events. The practical measure is whether operators still respond after a month of operation.
Can false alarms be eliminated entirely?
Only at the cost of misses. Setting thresholds is always a trade-off between the two, made deliberately for each scenario.
Does the number of alarms fall over time?
Usually yes, through tuning in the first weeks and through changes in work organisation that the system helps reveal.