Guaranteeing Human Privacy

This topic presents an internship opportunity focused on developing privacy-preserving AI techniques for camera-based systems used in industrial environments. The project aims to ensure human privacy by anonymizing identifiable attributes in videos while retaining useful data for analysis.

• Privacy-preserving AI model: An auto-encoder based model has been developed by IDLab Ghent and Flanders Make to hide identifiable human features such as tattoos and glasses in videos, enabling downstream algorithms to still extract key insights like human 2D keypoints. This model is integrated into a two-stage demonstrator involving segmentation and filtering of images.

• Internship objectives: The internship involves designing a robust toolbox from the existing demonstrator, implementing anonymization metrics, benchmarking real-time segmentation models, testing different camera models, investigating human re-identification on filtered images, and pre-training the auto-encoder with simulation data validated on real-world data.