Computer vision training data collection process

 

Why are photos collected?

With PeopleCounter, bypassing persons are detected instantly within the mobile device, eliminating the need to transfer or store data on servers for analysis. Video is not streamed, captured, or stored. No biometric data is collected, stored, or processed at any point in the entire process. No photographs are stored or captured. The system only detects the human figure on the device, and only numerical data is collected. We are now further developing our person recognition algorithm for our commercial service to be even more accurate. To do this, we need photos of people to train a more accurate algorithm so that no photos need to be stored in the ready commercial product.

What happens to the collected photos?

Photos are collected and stored locally in our office. Nothing is moved outside the office at this stage. Video recordings are transferred from two phones to computers locally with a cable. After that, they are split into individual photos, and then reviewed by our team. Original videos are now deleted. Personal data, i.e. photos where persons can be identified are anonymised by blurring the face. As a tip, feel free to wear a hat at our party and thus your face will not be recognized at all, because the video is recorded from above. After anonymization, generative AI is used to create further variations of some of the photos. Finally, once all the material has been carefully prepared, the training material is processed on the most powerful supercomputer in Europe, LUMI, in Finland. After that, we have a newer, more powerful, and more accurate algorithm providing 100% privacy-preserving automation for people counting