Scan car license plate recognition

Municipality of The Hague

The purpose of the algorithm is to automatically scan license plates for the purpose of parking enforcement.

Last change on 30th of June 2025, at 9:29 (CET) | Publication Standard 1.0

• Publication category: Impactful algorithms
• Impact assessment: DPIA
• Status: In use

General information

Theme

Traffic

Begin date

2021-06

Contact information

[e-mail]

Link to publication website

https://www.denhaag.nl/nl/parkeren/privacyverklaring-parkeercontrole-met-scanvoertuigen/

Responsible use

Goal and impact

Parking enforcement by scanning license plates after which it is determined whether the vehicle had a parking right. Those vehicles that do not have a valid parking right will be issued an additional charge.

Considerations

A manual check of all license plates is less accurate and takes much more time and therefore tax money.

Human intervention

All license plates read by the algorithm are manually checked before an additional charge is imposed. This is done by two people independently. The blurring algorithm is activated only after a human review has taken place, such as when dealing with an objection. The outcome of the algorithm - the blurred image - is then checked by an employee before it is sent to the citizen. If necessary, the image can be manually adjusted or still declared unfit for inspection.

Risk management

During procurement, a supplier was chosen that complied with all laws and regulations and could deliver the desired level of security and information security that the municipality required. When deploying the blurring algorithm, the main risk is that the algorithm incorrectly detects faces or insufficiently anonymises them. To mitigate this, use of the algorithm was limited to specific situations where a citizen objects or requests inspection. In addition, the algorithm has been tested for reliability in different lighting and imaging conditions. Any updates or changes to the software are evaluated in advance. A solution has been chosen that is technically separate from the primary license plate algorithm, to avoid mixing functions.

Legal basis

Tax legislation combined with traffic orders by the Municipal Executive stipulating which streets in the municipality will have paid parking.

Impact assessment

Data Protection Impact Assessment (DPIA)

Operations

Data

License plates of parked vehicles in paid parking areas.

Technical design

The primary algorithm processes images from scanning cars to detect license plates and compare them with data in the National Parking Register (NPR). This is done via Optical Character Recognition (OCR), a technique that automatically extracts the letters and numbers of a number plate from an image. The algorithm identifies relevant regions in the image, recognises the characters and translates them into a text format. A database comparison with registered parking permits then takes place. The OCR algorithm is not self-learning, but uses pre-trained models for character recognition.

In some cases, a second algorithm is used specifically for privacy protection. This algorithm detects faces and renders them unrecognisable through automatic blurring. This process is activated only if a citizen objects to an aftercharge and/or requests access to the image recording. The blurring functionality is a separate software application and is both procedurally and technically separate from the license plate recognition algorithm. The face recognition and blurring may make use of neural networks, and additional documentation can be added if needed.

External provider

ARS Traffic & Transport