Capability
False-alarm filtering — the feature that makes every other feature usable
CCTV false alarm reduction sounds like a minor optimisation until you realise it is the reason most security systems end up switched off. A system that sends forty alerts a night trains you to ignore it within a fortnight, and an ignored system protects nothing. PGAK classifies what it sees before it decides to interrupt you.
How cctv false alarm reduction works
Classify before alerting
Every moving object is identified as a person, vehicle, animal or environmental noise. Only classes you've asked about can raise an alert.
Apply zone and schedule
A person in the yard at 2am matters. The same person in the same yard at 2pm during dispatch does not. Time and place are part of the decision, not an afterthought.
Suppress known people
Enrolled faces pass silently. This alone removes most of the daytime noise at any site with staff.
Tune over the first fortnight
Every site has its own quirks — a streetlight, a neighbour's dog, a flapping tarpaulin. We tune thresholds against your real footage rather than shipping a generic default.
Where it earns its keep
- Perimeter alarms that were disabled because they fired all night
- Home camera apps whose notifications were muted months ago
- Warehouse yards with stray animals and constant vehicle movement
- Any site where a guard has learned to ignore the buzzer
What it won’t do
Every AI vendor lists strengths. These are the limits, so you can plan around them instead of discovering them.
- Filtering trades a small amount of recall for a large amount of precision. Tuned aggressively, it will occasionally suppress a genuine but ambiguous event — we set that balance with you rather than for you.
- Cameras pointed at a public road will always see more legitimate movement; the fix is framing, not filtering.
- Insects on the lens at night are the hardest single case, and are handled better by an IR housing than by software.
False-alarm filtering — common questions
How much can false alerts actually be reduced?
Sites typically see a 90%+ reduction after the first fortnight of tuning. The bigger change is qualitative: alerts go from something you swipe away to something you look at.
Why does my current system alert for shadows and rain?
Because conventional motion detection compares pixel changes between frames and has no concept of what an object is. Rain, shadows, headlights and swaying branches all change pixels. Object classification is a fundamentally different approach.
Could filtering cause me to miss a real intrusion?
It's a real trade-off and we won't pretend otherwise. Filtering set too aggressively can suppress an ambiguous event. We start conservative, review the first weeks of events with you, and tighten only where the footage justifies it.