AI Fisheye Dewarping & Analytics in Abu Dhabi

A fisheye lens projects the hemisphere below the camera onto the sensor with a known mathematical distortion model — typically an equidistant or equisolid projection. The dewarp engine inverts that model: for every pixel in the rectilinear output it computes the corresponding pixel on the fisheye sensor with bilinear interpolation. The result is a normal-looking view. The camera can emit several dewarp outputs simultaneously over RTSP: a panorama strip that flattens the perimeter of the circle, a dual panorama for ceiling mounts in a corridor layout, four virtual PTZ windows that an operator can pan and zoom independently, and the raw fisheye stream for evidential retention. Each rectilinear view feeds a detector trained on rectilinear imagery so the model's geometry assumptions hold. People-counting analytics overlay a tripwire on the dewarp output and count crossings; dwell-time analytics compute per-zone time-in-zone with a polygonal mask; heatmap analytics aggregate person-positions over a window of minutes or hours. Abu Dhabi specifics matter. Glossy floor tiles in malls and hotel lobbies create specular reflections that fool naive people detectors — the analytics must use the head-and-shoulders region, not the floor projection. Glass walls in towers cause the fisheye to see through to the corridor beyond, doubling up tracks if the zone mask is not drawn tightly. Ceiling height matters too: a six-metre mall ceiling needs a higher-resolution sensor than a three-metre clinic ceiling for the same pixel-on-target. Emirtech surveys the site, sizes the sensor against the longest required target range, sets the dewarp configuration, draws the analytics zones against the actual floor plan, and signs off only after checking detector accuracy on a representative traffic sample rather than a vendor demo loop.