AI Vehicle Make, Model & Colour Recognition for Abu Dhabi

The mechanism is a classifier — usually a convolutional or transformer-based image model — running on the cropped vehicle image from the same detector that feeds the plate pipeline. The output is a structured tag set: body type (sedan, SUV, hatchback, pickup, van, bus, truck), make and model class with confidence, and dominant colour with a colour-distance score against a palette. Some engines add a wheel-position and badge-position cue that adds discrimination between same-platform models that share a silhouette. In Abu Dhabi the realities that shape the install are precise. The vehicle mix leans heavily white, with strong representation of SUVs and high-end sedans that share silhouettes across model years; colour alone is weak, model classifier confidence matters. Strong sunlight on the bonnet flattens the colour reading unless the camera has wide-dynamic-range capability. Night-time reads need the same IR illuminator that ANPR uses, but the MMCR colour channel collapses under IR; we hold the colour tag at a lower confidence and only commit to model and body-type tags at night. The system also rejects low-confidence tags rather than guess — a documented "unknown" is more useful in a forensic search than a wrong make tag. Emirtech specifies the sensor, calibrates the classifier confidence threshold, and runs an acceptance test against a known fleet to document the measured tag rates across day and night cycles, by body type and by colour.