Cutting false alarms 78% for a Dubai CCTV monitoring SOC
Dubai-based CCTV monitoring SOC ·
−78%
False alarms
+40%
Cameras per operator
−55%
Mean response time
Challenge
Operators were drowning in 4,000+ daily motion alerts, 92% of which were false (wind, shadows, animals).
Approach
Edge inference on 240 cameras with a YOLO-based human/vehicle classifier, behaviour-aware filtering, and a fall-back to a central VLM for ambiguous events. Tuned thresholds on six weeks of site data.
Outcome
Daily alerts dropped to ~860, with operator confidence restored within two weeks. The SOC now handles 1.4× the camera count without adding headcount.