AI Dwell-Time & Zone Analytics for Abu Dhabi Floors

The mechanism is a person detector paired with a multi-object tracker. The detector finds each person above its confidence threshold inside one or more zone polygons; the tracker assigns a stable identity (a track ID) to each detection across frames using motion prediction and a simple appearance signature — bounding-box IOU between consecutive frames, plus a colour or feature embedding to recover from short occlusions. The dwell timer starts when a track enters a zone and persists past a minimum-time floor — typically three to five seconds — that filters out simple pass-through. The output is a clean per-zone metric: number of dwell events, mean and 95th-percentile dwell time, and the time-of-day curve. In Abu Dhabi specifically, three realities shape the install. First, mall fixtures and showroom displays are high-gloss; reflections create ghost detections that a generic tracker treats as new people, inflating the count. The fix is mounting angle plus a tuned appearance model. Second, abaya and dishdash silhouettes confuse short-occlusion re-identification when two shoppers cross paths — we tune the tracker's IOU and embedding parameters against your own footage rather than accept defaults. Third, public-realm cafe zones along the Corniche or in mall food halls have people sitting for very different reasons; the dwell floor and zone polygon must be calibrated to the operational question — is this a queue zone, a browse zone, or a sit-and-rest zone? Emirtech specifies, calibrates, and signs off only after an acceptance test that compares automated dwell against a stopwatch sample.