AI Drone Detection on CCTV — Honest, Detection-Only Scope

A drone-detection camera is, mechanically, a normal IP camera — but the classifier on the edge or on a dedicated server is trained on small aerial objects against sky and built-environment backgrounds. The model looks for a tight cluster of pixels with a characteristic motion signature: smoother trajectory than a bird, persistent against wind drift, often with rotor flicker at higher zoom. PTZ cameras can be slaved to a wide-angle 'spotter' so that when the spotter classifies a candidate, the PTZ auto-zooms for a confirmation crop. In this niche the false-positive failure modes are very specific. Sea birds banking near the Corniche, kites over Saadiyat, plastic bags lifted on a thermal off Mussafah asphalt, distant helicopters on approach to Al Bateen, and floodlight reflections on humid nights will all flicker into a naive classifier. A serious deployment tunes confidence thresholds per camera, blacklists known flight corridors and rooftops, and demands a multi-frame persistence check before raising an alarm. Emirtech specifies the camera count and placement using the actual sky cones the client cares about — not a uniform grid — installs them with sky-aware mounts that minimise sun glare, and tunes the classifier on-site over an initial tuning period using captured footage of real bird traffic from that exact location. We integrate the alert into the existing VMS so the operator sees a video pop-up and audit log entry, with the option to push to a duty mobile number. We do not connect any active counter-measure.