People can be unintentionally captured by camera-based indoor positioning systems, raising privacy concerns. The effects of anonymising these images on positioning performance and the preservation of scene information were evaluated.
Face-level, person-level and scene-level anonymisation methods were compared using person detection recall, semantic object retention and indoor zone classification. Among the evaluated methods, face swapping best preserved scene information and positioning performance. When the entire person needs to be removed, YOLO-based person inpainting offers a useful trade-off, retaining more scene information than global blurring.
See more:
- Martina Kuchtová, and Miroslav Opiela. Impact of Visual Data Anonymisation on Vision-Based Indoor Positioning (IPIN 2026 conference, WCAL paper)
- Martina Kuchtová. Enhancing Camera-based Indoor Positioning with Neural Networks, 2025, master thesis,
- Poster