Please note: The algorithm descriptions in English have been automatically translated. Errors may have been introduced in this process. For the original descriptions, go to the Dutch version of the Algorithm Register.
Fixi - Notifications Public Space
- Publication category
- Impactful algorithms
- Impact assessment
- Field not filled in.
- Status
- In use
General information
Theme
- Space and Infrastructure
- Public Order and Safety
Begin date
End date
Contact information
Link to publication website
Link to source registration
Responsible use
Goal and impact
If something needs to be fixed or cleared in the street or in a park, this can be reported to the municipality via the Fixi app, the municipality of Arnhem's notification system. A dangerous traffic situation, housing nuisance or nuisance from people and catering establishments can also be reported. The notifier himself chooses which main category fits the report best (e.g. 'nuisance' or 'street furniture'). Employees of the municipality check the submitted reports and divide them into sub-categories, so that the report reaches the right municipal handling team. Associated with each (sub-)category is a predetermined handling period.
Considerations
The use of the algorithm in Fixi makes processing public space reports faster, more accurate and clearer. Reports are automatically recognised and classified, allowing the right department to get to work faster. This allows the municipality to work more efficiently and reports are followed up better.
A possible drawback is that the report may be accompanied by photos showing persons, license plates or other personal data. To limit this risk, the algorithm automatically recognises these data and makes them unreadable (blur). Thus, residents' privacy remains protected.
Other ways of processing, such as fully manual assessment of reports, would take more time and capacity and lead to longer processing times. The deployment of this algorithm is therefore considered reasonable and justified. Residents also retain control, as they decide whether to share personal data when submitting a report.
Human intervention
Reports from Fixi arrive at the work distributors. These reports are already anonymised automatically. An employee then checks the report and, if necessary, performs additional or corrective anonymisation. Based on the notification, the work distributors then decide what to do and who performs the task. So there is always human intervention in the checking and further processing of notifications.
Risk management
The algorithm has no impact on a decision, but is designed to make any personal data unrecognisable in photos. Municipal employees see the original photos and dates. Residents see the censored photos. The algorithm may not properly recognise a personal data, this will be acted upon on a situation-by-situation basis by the coordinator.
Legal basis
- Municipal Act
- Roads Act
- Environmental legislation
- General local government act (APV)
- Licensing and Catering Act
- Noise Pollution Act
- Soil Protection Act
- Housing Act
- Town and Country Planning Act
- Nature Conservation Act
- Flora and Fauna Act
- Attractions and Games Equipment Act Decree
Link to Processing Index
Elaboration on impact assessments
Not applicable
Operations
Data
The algorithm in the Fixi app processes data on reports in public spaces. This data is mainly about the type of report (such as broken lights or litter), the location and when the report was made. The information is used to automatically categorise reports and forward them to the right department or employee.
In addition, the submitter can choose to also provide personal data, such as name, address (not mandatory), phone number and/or e-mail address. These data are only used to contact you about the report if necessary.
It is up to the submitter whether these personal data are included: a notification can also be filed without sharing these data. Once a report is submitted, all personal data are automatically anonymised before the algorithm processes the report further. A staff member checks this and performs additional anonymisation if necessary.
Technical design
Number plate recognition is performed with an ALPR module that combines OpenCV's edge detection techniques with Pytesseract's LSTM-based OCR engine for accurate character extraction.
Face detection is performed with the RetinaFace library, which uses a deep learning-based single-shot detection algorithm optimised for localising facial features.
For text detection, we use EasyOCR, which uses the CRAFT algorithm for identifying text regions and CRNN for recognising text content.
To ensure privacy, sensitive areas such as faces, number plates and specific text elements are blurred:
- Gaussian convolution for image smoothing,
- Colour space transformations for preprocessing,
- and bitmap rendering for annotations such as rectangles and labels.
For text masking, we use a Named Entity Recognition (NER) module with the BERT algorithm, which provides contextual understanding for identifying sensitive textual entities. These entities are then masked using regular expression-based pattern matching algorithms.