AI Queue-Length Management for Abu Dhabi Counters & Tills

The mechanism is a queue-polygon detector running on the CCTV feed of the queueing corridor. The polygon is drawn over the actual queue lane — single-line or multi-line, depending on the rope-and-post layout. Inside that polygon, a head-and-shoulders classifier picks each person above its confidence threshold; the per-frame count is smoothed temporally so that a person glancing at their phone and stepping sideways does not get double-counted. A second model estimates wait time, either by tracking each detection's residence time inside the polygon, or by dividing instantaneous occupancy by an observed throughput rate. Operational signals are then generated against thresholds you set with the floor manager: a soft warning when the queue exceeds, say, five people, and a hard warning when it exceeds the agreed service-level number that should never be hit. In Abu Dhabi specifically, the realities that break naive installs are these. First, queue corridors in customer-service halls are often glass-fronted and lit from one side, putting half the queue in shadow at certain hours; the lens choice must keep usable pixel density across the whole polygon. Second, customers carry briefcases and shopping bags that defeat low-angle detectors — we mount higher, more overhead. Third, groups, children and prams create occlusion, which means the classifier confidence threshold needs to be tuned on your own footage, not on a default. Emirtech agrees an acceptance test with you, comparing AI count to a manual clicker, and documents the resulting error band as the basis for the alerts.