Supplementary Material for: Impact of Visual Data Anonymisation on Vision-Based Indoor Positioning

This repository contains a demonstration and replication kit to support the research paper Impact of Visual Data Anonymisation on Vision-Based Indoor Positioning. Although the primary study was conducted using an extensive private dataset of indoor environments (which is not disclosed due to GDPR and biometric privacy constraints), this repository provides synthetic samples and detailed detection logs which are intended to validate the proposed anonymisation methodology. This repository contains: anonymisation_and_detection_validation_data demonstration_videos: Three video sequences from an independent building zone demonstrate the visual fidelity of various anonymisation techniques. These include person-centric methods (inpainting, blurring, pixelation, blackout, and face swap) as well as scene-level protection, which involves blurring the entire scene. The folder contains both raw and anonymised sequences, allowing for straightforward direct comparison. detection_logs: Frame-by-frame detection logs (CSV) for the demonstration videos were used to document detection continuity and stability. MOT15_validation: Frame-by-frame performance analysis (CSV) comparing the accuracy of YOLO- and Haar-based person detection models using the MOT15 training sequences. anonymisation_methods: Jupyter Notebook demonstrating the implementation and application of the various anonymisation techniques discussed in the paper. key_object_recognition_weights: YOLO_vips_key_objects.pt: fine-tuned YOLO weights specifically optimised for Key Object Recognition in indoor environments. class_mapping.txt: reference file defining the classes trained into the model.