Automated parking barrier gate raised at a vehicle entrance

How ANPR Works

Automatic Number Plate Recognition — the engine that reads a plate, corrects it, and decides whether to open the gate.

What happens between a car arriving and a gate opening

Automatic Number Plate Recognition (ANPR, also called ALPR) is often described as though a camera simply reads a plate and a barrier lifts. In a real car park it is less direct than that. Plates are curved, split across two rows, covered in snow, washed out by low sun, or held on with black screws that a camera reads as letters.

This page describes what Autlo actually does with a plate between the moment a vehicle arrives and the moment a gate opens — including the cases where recognition reaches its limit and what happens then. Autlo has been building this since 2013, and the examples below are from live sites.

Works on top of ANPR raw results

Autlo is software. The recognition can come from a minicomputer on site running Autlo's own engine, or straight from a camera that already does its own recognition. Either way, Autlo's logic runs on top of the raw result.

Autlo access camera big2 1 2 80a35ecd7b

Autlo ANPR engine

Runs on a minicomputer on site. It takes the camera feed and applies Autlo's own reading and correction logic.

Tech anpr camera traffic f5047afe66

The camera's internal ANPR

Where the camera already recognises plates well, Autlo connects over its API and runs the parking operator and access management logic on those results instead.

Understanding and correcting the plate

Raw recognition slips on characters that look alike. Country plate formats are what resolve most of it.

The problem

The engine can hesitate between characters that share a shape: O and 0, 1 and I, 2 and Z, 8 and B.

The correction

Standard plates follow known country formats. Autlo adjusts the decision to fit the format that applies at that site.

In practice

Sweden, Finland and Lithuania use ABC123; Estonia uses 123ABC. So a read of 7I0AK0 resolves to 710AK0, and an ambiguous 01Z0I2 resolves differently in Sweden than in Estonia. The country decides.

Alternate plates

Country formats handle the ordinary cases. Special plates need something else: rather than overwriting what the recognition believes, Autlo keeps that reading and adds the likely candidates alongside it as alternate plates. The vehicle is then matched whichever way it was read.

BMWI8

A personalised plate that matches no local standard. Is the character an I or a 1? The engine often reads BMWI8, sometimes BMW18. Both are stored, so the vehicle is detected either way.

LCC999K

A curved plate split across two rows, with reflections. The engine often returns only LCC or only 999K. Both fragments are kept as alternates, and the gate still opens.

PJ59TDX

Black screws sitting inside the characters. The engine reads PJ59T8X, because a screw closes the D into an 8. Both readings are stored.

How the software understands context

A camera produces many reads of the same vehicle. Turning that stream into one correct parking session is where most real-world errors are prevented.

The gate opens once

One vehicle usually produces the same read several times in a row. A time window ignores the repeats, so a session starts only once. The default window is 30 seconds.

On exit, vehicles that just entered are ignored

A back-plate caught by the entry camera should not end a session that just began. After entering, a vehicle is ignored at the exit for a set window.

On entry, vehicles that just exited are ignored

The mirror rule. After leaving, a vehicle is ignored at the entrance for a set window, so a back-plate read on the way out cannot start a new session.

Direction rules per gate

Each point is told which way traffic flows — coming, going, or stationary — so entries and exits are never mixed up.

Getting the best read in the field

Recognition quality is decided at installation more than in software. These are the things that make the difference on site.

Mask unrelated areas

Mask out everything that is not the lane you care about. Ideally the camera never sees passing traffic or vehicles heading elsewhere.

Light at night

A simple LED bar above the lane makes a clear difference after dark, and costs very little to add.

Camera angle

The straighter the view onto the plate, the better. Sharp angles cost accuracy before any software is involved.

Camera height

Mounting higher than the vehicle's front lights greatly reduces reflections when the beam points at the lens.

Sun reflections

At certain angles a plate washes out fully white for part of the day. Plan for it rather than discovering it in June.

Snow fallback

Decide the fallback before winter: the app, front and back cameras, or no enforcement during snow.

If you can read the plate, the software can read it too. If a human cannot, neither can we.

That is the honest limit of the technology, and it is the most useful thing to know before planning a site. Everything above exists to get as close to that line as possible. Nothing gets past it.

Where ANPR reaches its limits

These are real cases from live sites. Each one is a reason to have a backup rather than a reason to avoid ANPR.

Water on the lens

Rain sitting on the glass breaks the focus and scatters light from headlamps across the frame.

The plate is blocked

A bike rack, a tow bar or a trailer physically covering the plate. No camera reads what is not visible.

Sun

Low sun at the wrong angle washes the plate to flat white, or throws the vehicle into silhouette.

Snow

Snow and road salt packed across the plate after a winter drive. Front and back cameras help here.

Digital noise

Compression artefacts in the video stream, usually a network or camera configuration problem rather than a recognition one.

Not a vehicle at all

Occasionally the engine finds text where there is no plate. The lettering on a security guard's jacket has been read as a plate more than once.

What happens when a read fails

A site should never depend on a single successful read. There are four layers behind it.

Constant evaluation

If a read fails once, the driver usually waits. Autlo keeps evaluating for as long as the vehicle is visible — in a gated area, until the gate opens.

Several cameras

Two sides, or front and back. Front and back keeps working when snow covers the front plate, and it catches bikes, mopeds and trailers.

App backup

Drivers open the gate from the Autlo app whenever it is needed, without calling anyone.

Helpdesk and back office

Remote opening from the helpdesk and back office as a final fallback.

Speed, monitoring and what the driver sees

A session starts in seconds

Arrival, processing, data sent, data received, session started — the whole sequence typically completes within a couple of seconds of the vehicle arriving. Recognition quality is measured and monitored continuously rather than assumed.

Monitoring and failover

Autlo monitors that every camera is online. If one recognition server fails, another machine takes over immediately.

The driver can see the result

The recognition result can be shown on a screen at the gate, with the plate and the free spaces remaining, so the driver knows exactly what the system decided.

Number plate data and GDPR

All servers in the EU. The main server is hosted in the Netherlands. The recognition server can sit on site, in Autlo's office, or in yours.

Images are deleted automatically. Images are stored for four months and then removed, with no manual cleanup step to forget about.

Running in the field

  • Merko edf2f3ef01
  • 8999 swedbank logo default 1 bd0f872674
  • Screenshot 2026 01 30 at 13 46 32 7eb2b4aea3
  • 15minlogo 064ec658ad
  • Chroma gif pagespeed ce T1 Gh KZR 2w E 9be39af8bd
  • Avalikud teenused logo 615877f999

Common questions

What accuracy should we expect?

The honest answer is the one above: if a person looking at the camera image can read the plate, the software can read it too, and if a person cannot, neither can the software. Accuracy at a given site is therefore decided mostly by camera placement, lighting and angle rather than by the recognition engine. Autlo monitors read quality continuously so the number is measured rather than estimated.

Can we use the cameras we already have?

Yes. Autlo runs on top of the feed from cameras already installed, and is not tied to one manufacturer. Where a camera already performs its own recognition well, Autlo connects over its API and uses those results instead of duplicating the work.

What happens when a plate is covered in snow or dirt?

Recognition fails, as it would for a person. That is why sites are designed with a fallback: front and back cameras so a clean rear plate still works, the app for drivers to open the gate themselves, and remote opening from the helpdesk. Before winter it is worth deciding explicitly which fallback applies during snow.

Does it work at night?

Yes, with adequate light. A simple LED bar above the lane makes a clear difference after dark and is inexpensive to add during installation.

What if a camera or the recognition server fails?

Every camera is monitored for availability. If one recognition server fails, another machine takes over immediately, and the gate can still be opened from the app or by the helpdesk.

Does it detect motorcycles, mopeds and trailers?

A second camera covering the rear of the vehicle is what makes this reliable, since many of these carry a plate only at the back.

Where is plate data stored and for how long?

All servers are in the EU, with the main server in the Netherlands. The recognition server can sit on site, in Autlo's office, or in yours. Images are kept for four months and then deleted automatically.

How does the system avoid opening the gate twice for one car?

A time window, 30 seconds by default, ignores repeated reads of the same vehicle, so one arrival produces one session. Matching rules also stop a back-plate read at entry from ending a session, and a back-plate read at exit from starting one.

Planning a site?

If you are specifying cameras, deciding where to mount them, or working out what happens on the days recognition fails, we are happy to go through it with you before anything is ordered.

Questions about plate recognition?

Tell us about the site — the gates, the cameras you have, and what should happen when a read fails. We will tell you what is realistic.

Book a Demo