Because reading a plate on a moving vehicle needs a fast shutter and dedicated infrared illumination, and a general camera has neither.

A normal exposure smears the plate on a moving vehicle into an unreadable blur.

A fast shutter freezes it and admits very little light, so the image would be almost black without dedicated illumination.

ANPR cameras use infrared illuminators tuned to exploit the retroreflective plate coating, so the plate returns far more light than the surrounding vehicle.

The result looks strange to a human - a bright plate on a dark vehicle - and is exactly what the recognition software needs.

That is why an ANPR installation normally has two cameras: one for the plate and one for a conventional overview image of the vehicle.

As square to the plate as possible, within the manufacturer's stated horizontal and vertical angle limits - the geometry decides the read rate.

A plate viewed at a steep horizontal angle is compressed and cannot be read reliably.

Mounting too high produces a steep vertical angle with the same effect, which is the commonest error on gantry and pole installations.

Manufacturers publish maximum angles, and installations outside them do not recover through software tuning.

The camera also needs to be positioned so the plate is a specific size in the frame at the read point, which sets the lens and the distance.

Survey the lane properly, including where vehicles actually stop, which is often not where the design assumed.

High in a controlled lane with good geometry; substantially lower on open roads with multiple lanes, speed and weather.

A single-lane entrance where vehicles slow or stop, with correct geometry, is the easiest case and performs very well.

Multiple lanes, higher speeds, motorcycles, obscured or damaged plates, and non-standard plate formats all reduce it.

Dirt, snow and towbars obscure plates, and there is nothing the camera can do about that.

International traffic introduces plate formats the software may not be configured for, which needs specifying rather than assuming.

Ask for read rate figures measured in a comparable installation rather than in ideal conditions, and agree how it will be measured on acceptance.

As an ongoing operational process - the system is only as useful as the currency of its lists.

Plates are matched against permitted, barred and expected lists, and each has an owner and a maintenance routine.

A permit list that is not updated when staff and contractors change produces both refused legitimate vehicles and admitted former ones.

Integration with the source of truth - a staff system, a booking system, a visitor system - is what keeps it current without manual effort.

Expiry dates on entries are essential, particularly for contractors and visitors.

Barred lists raise their own questions about who may add an entry, on what basis, and how long it stays - which are policy matters as much as technical ones.

Number plates are personal data in most privacy regimes, so purpose, retention, access and signage all need to be settled locally before deployment.

An ANPR system records where a vehicle was and when, which is movement data about identifiable people.

Most regimes require a stated and legitimate purpose, a defined retention period, and controls over who can access the records.

Several require notification signage where the system operates, and some require a formal impact assessment before deployment.

Automated enforcement - charging or penalising on the basis of a read - is regulated separately again and is not permitted everywhere.

This differs enough between territories that a policy written for one country should not be assumed to work in another.