Low-Light Colour AI CCTV Cameras in Abu Dhabi

The mechanism is sensor physics, not marketing. A larger 1/1.8" back-illuminated CMOS sensor collects roughly two and a half times the photons per pixel of a typical 1/2.8" sensor at the same resolution, and an F1.0 aperture lets in roughly twice the light of an F1.4 lens. Combined with a low read-noise CMOS readout and a colour-aware noise-reduction pipeline, the camera holds a usable colour image at illuminance levels where a standard camera has already cut to infrared. That matters for AI because most detection models — person, vehicle, two-wheeler — are trained on colour imagery; running them on noisy monochrome scenes raises the false-negative rate and degrades AcuSense or WizMind class motion-filter accuracy. The supporting layer is the codec: an H.265+ smart encoder cuts bitrate aggressively on quiet zones, but if its quality floor is too low it smears the small edges the detector needs at the back of the scene, so the AI region gets a minimum bitrate budget regardless of motion. Abu Dhabi specifics push this further. Streetlight spill at compound boundaries is uneven, so the camera must hold colour at the dimmest patch of the scene, not the brightest. Sandstorms scatter light and add a global haze that confuses the auto-exposure, and we tune the gain ceiling so the ISP does not push the read noise into the AI region during a dust event. Mixed sodium-vapour and LED street lighting produces a colour cast the white-balance routine has to handle without crushing skin tones or numberplates, and the IR cut-off threshold is set against measured site illuminance rather than the factory default. Emirtech selects the lens, configures the codec, walks the site at night, and validates that the analytics still fire correctly under the real lighting your guards will see at 02:00.