Image recognition advertising tax

Municipality of Amsterdam

Amsterdam wants a quieter streetscape and therefore wants to levy an advertising tax. Advertising tax must be paid for displays visible from the public road (such as façade advertisements, signs and window advertisements).

Last change on 9th of December 2024, at 14:28 (CET) | Publication Standard 1.0

• Publication category: Impactful algorithms
• Impact assessment: Field not filled in.
• Status: In use

General information

Theme

Work

Begin date

Field not filled in.

Contact information

[e-mail]

Responsible use

Goal and impact

Amsterdam wants a quieter streetscape and therefore wants to levy an advertising tax. Advertising tax must be paid for expressions visible from the public highway (such as façade advertising, signs and window advertising). The size of the advertisement, in addition to the tariff (area) and the type of display, among other things, determines the amount to be paid. To collect the taxes, scan cars drive on the streets several times a year. The cameras in the scan cars recognise: advertisements; provide them with coordinates; with a date; measure the surface area; personal data such as faces and license plates are blurred and not stored. The system then makes a proposal to the employee for the advertisement. The clerk can see the image (photo) taken and the data captured. The clerk assesses this and determines the taxpayer and tax liability (type of advertisement and area). The system learns from this so that the cameras better recognise and measure the advertisement next time on the street. Before the assessment is made, the taxpayer can check in the digital counter of Taxes whether everything is correct and pass on changes. Then the tax assessment is sent and the Advertising Tax can be paid.

Link to service described:

https://www.amsterdam.nl/belastingen-heffingen/

Considerations

Field not filled in.

Human intervention

The system only makes a proposal for the advertisement to the clerk. The clerk always decides. The taxpayer has the opportunity to check the advertisements placed on assessment and submit amendments to them. The possibility of objecting and appealing against the assessment exists. This can be done at the digital desk of Taxes.

Risk management

The system is not risky in terms of privacy. It does not record personal data and only takes pictures of displays in public spaces. The biggest risk here is that the location of the advertisement becomes known, which could possibly make it possible to trace the address of a person or company. This is also possible on the street. Unauthorised persons cannot access the stored images due to the realised information security. Security risks are controlled by the overall security set-up on systems and connections. These meet the requirements. In execution, the biggest risk is that an advertisement is not scanned or measured correctly, resulting in someone getting the wrong assessment. However, the clerk always checks and determines the tax liability. The taxpayer can respond to the recorded advertisement and can object and appeal the tax assessment. In implementation, the biggest risk is that an advertisement is not scanned or measured correctly, resulting in a person receiving the wrong assessment. However, the clerk always checks and determines the tax liability. The taxpayer can respond to the recorded advertisement and can object and appeal the tax assessment.

Operations

Data

Advertisement For each recognised advertisement in the public space, an image of the public space showing that advertisement is used (via the coordinates). Personal data are blurred and not recorded. Location data The camera and associated software record the location of the recognised advertisement. To this end, geographical coordinates are recorded for the advertisement using the basic topographical map of the Netherlands. Date The date of detection is recorded. Type The type of advertisement is automatically recognised and recorded. Measurement The camera and associated software record the dimensions of the advertisement. The image, date, location and measurement are sent to the tax application. In doing so, the system creates a new tax object or links this data to an existing advertising object. Tax confidentiality applies to this data. Feedback The results of the employee's assessment such as type of advertising object, location and measurements are automatically fed back to the system. The system learns from this. Tax data The location data is used to link the advertising object to the WOZ object (e.g. shop). The tax data of that WOZ object is used to determine the taxpayer and compile the tax liability. Through a regular process, tax assessments are made. Tax secrecy applies to this data.

Technical design

Architecture of the model: The algorithms are used to: Recognise a possible advertisement and capture it as an image. We adopt Cyclomedia's service which is widely used to capture public space. Patented positioning technology is used. This technology is not made public because it is company information. In the imagery, the system looks for the advertisements and records them in coordinates. This data is used to determine the tax object. The accuracy in the recognised objects is at least 95%. Through the learning effect, this percentage is being increased. In addition, there is a control process that uses file analysis to check whether there are tax objects (e.g. restaurant) where advertising is expected but for which no tax liability has been created. The location data of the advertising is recorded. This data is used to determine the taxpayer. To determine and record the type and dimensions of the advertisement. These data are used to determine the basis of taxation. Record the date of the observation. This data is used to determine the levy period. To learn from the data confirmed by the clerk. This data is used to further optimise image recognition. Performance Photogrammetry is involved. Photogrammetry deals with the interpretation and measurement of imagery in determining and describing the shape and size and location of objects. Here, the shape, location and dimensions are determined using mathematical models within an automated algorithm. The mathematical models used are not publicly available. This is confidential business information. No (inter)national standards for terrestrial positioning, photogrammetric and point cloud measurements are available. It is the employee who confirms the location and calculates the area of the advertisement. The rules governing this are set out in the tax ordinance. The employee's application of these rules thus determines the accuracy and precision of the measurement. The algorithm is self-learning. It uses the results of the employee's inspection to learn so that the image recognition becomes more accurate.