Protecting Privacy Of Citizens With Personal Data Anonymization Service

One of the many challenges related to the usage of Computer Vision based solutions in many real-world scenarios is to ensure the personal data protection of the people who might be affected by them. For a Smart Bicycle Highway from the Lahti Travel Centre to Ajokatu currently built by the CitiCAP project this matter is especially important. The large crowds of people moving around requires considering how to enhance their everyday life experience with smart solutions and services without intruding into their privacy.  

Picture 1. Smart solutions aim to improve cycling experience in Lahti (CitiCAP 2020)

Personal Data Anonymization service is one of many services that was being prototyped and tested for the Smart Bicycle Highway at LAB University of Applied Sciences in 2020 (City of Lahti 2020). The purpose of this service is to detect and anonymize any personal or privacy sensitive data that can be found in images or videos and hide it from other services that will access this data later.  

How does it work? 

The implementation is based on the Azure Cloud Platform and uses Event Driven Architecture to track incoming images and anonymize them in real time. The AI model created with the Azure Cognitive Services analyzes each image separately, identifies all objects that require anonymization and returns their coordinates to the service. The service will then apply a preferred anonymization filter to each identified object. Anonymized images are then stored in the cloud where they can be accessed by any other backend or frontend service. No other information is gathered or saved in the process. The service can receive data from many outer sources e.g. street cameras at the same time and return results for each source separately.   

Picture 2. Image after being processed by Anonymization Service (Image: Jevgeni Anttonen)

The model used for data anonymization was specifically trained to detect and identify any objects that can contain privacy sensitive information e.g. pedestrian faces and license plates. The selective and fully automated anonymization of specified objects is crucial in this case because the rest of the image will be likely used by other services for other purposes such as detection of road accidents or road surface analysis.  

Picture 3. Average precision for the model using 50% Probability Threshold (Image: Jevgeni Anttonen)

Working with the model and Azure Cognitive Services it was possible to reach an average of 90% of detection accuracy on the training dataset even with complex images with a large number of objects to recognize at the same time. The accuracy will also likely improve in the future as more new images and videos, especially ones from the street surroundings will be collected and used to enhance the existing dataset. 

Current state and further goals 

The first prototype version of the Personal Data Anonymization service was developed and tested in October 2020 at LAB University of Applied Sciences. It was also successfully presented during the CitiCAP steering group meeting. The service and the model are currently under development and will likely see significant improvements before being tested in a real street environment.   


Jevgeni Anttonen works as a developer and researcher in the CitiCAP project at LAB University of Applied Sciences. 


City of Lahti. 2020. CitiCAP-project. [Cited 26 November 2020]. Available at: 


City of Lahti. 2020. CitiCAP-project. [Cited 26 November 2020]. Available at: 

Uia-initiative. 2020. Lahti CitiCap – Citizen’s cap-and-trade co-created. [Cited 26 November 2020]. Available at: 

Microsoft. 2020. Microsoft Azure Overview. [Cited 12 May 2020]. Available at: 


Picture 1. CitiCAP. 2020. CitiCAP Facebook page. Available at: 

Leave a comment

Your email address will not be published. Required fields are marked *