Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adversary-Guided Motion Retargeting for Skeleton Anonymization

Thomas Carr, Depeng Xu, Aidong Lu

TL;DR

The paper tackles privacy risks in skeleton-based motion data by introducing PMR, a privacy-centric deep motion retargeting framework. PMR uses two encoders to separately capture motion and privacy information, a decoder to retarget motion onto a dummy skeleton, and embedding classifiers with a quality discriminator to balance utility and privacy. Through unpaired and paired training, including cross-reconstruction, triplet, latent-consistency, and end-effector losses, PMR achieves motion retargeting utility on par with state-of-the-art methods while substantially reducing privacy attack success. The results on NTU RGB+D 60 demonstrate effective disentanglement of motion and identity cues, offering a practical approach for privacy-preserving skeleton-based VR and analytics; future work may explore transformer-based architectures to further enhance representation learning and anonymization performance.

Abstract

Skeleton-based motion visualization is a rising field in computer vision, especially in the case of virtual reality (VR). With further advancements in human-pose estimation and skeleton extracting sensors, more and more applications that utilize skeleton data have come about. These skeletons may appear to be anonymous but they contain embedded personally identifiable information (PII). In this paper we present a new anonymization technique that is based on motion retargeting, utilizing adversary classifiers to further remove PII embedded in the skeleton. Motion retargeting is effective in anonymization as it transfers the movement of the user onto the a dummy skeleton. In doing so, any PII linked to the skeleton will be based on the dummy skeleton instead of the user we are protecting. We propose a Privacy-centric Deep Motion Retargeting model (PMR) which aims to further clear the retargeted skeleton of PII through adversarial learning. In our experiments, PMR achieves motion retargeting utility performance on par with state of the art models while also reducing the performance of privacy attacks.

Adversary-Guided Motion Retargeting for Skeleton Anonymization

TL;DR

The paper tackles privacy risks in skeleton-based motion data by introducing PMR, a privacy-centric deep motion retargeting framework. PMR uses two encoders to separately capture motion and privacy information, a decoder to retarget motion onto a dummy skeleton, and embedding classifiers with a quality discriminator to balance utility and privacy. Through unpaired and paired training, including cross-reconstruction, triplet, latent-consistency, and end-effector losses, PMR achieves motion retargeting utility on par with state-of-the-art methods while substantially reducing privacy attack success. The results on NTU RGB+D 60 demonstrate effective disentanglement of motion and identity cues, offering a practical approach for privacy-preserving skeleton-based VR and analytics; future work may explore transformer-based architectures to further enhance representation learning and anonymization performance.

Abstract

Skeleton-based motion visualization is a rising field in computer vision, especially in the case of virtual reality (VR). With further advancements in human-pose estimation and skeleton extracting sensors, more and more applications that utilize skeleton data have come about. These skeletons may appear to be anonymous but they contain embedded personally identifiable information (PII). In this paper we present a new anonymization technique that is based on motion retargeting, utilizing adversary classifiers to further remove PII embedded in the skeleton. Motion retargeting is effective in anonymization as it transfers the movement of the user onto the a dummy skeleton. In doing so, any PII linked to the skeleton will be based on the dummy skeleton instead of the user we are protecting. We propose a Privacy-centric Deep Motion Retargeting model (PMR) which aims to further clear the retargeted skeleton of PII through adversarial learning. In our experiments, PMR achieves motion retargeting utility performance on par with state of the art models while also reducing the performance of privacy attacks.
Paper Structure (29 sections, 16 equations, 10 figures, 6 tables)

This paper contains 29 sections, 16 equations, 10 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. Problem Statement
  5. Overview
  6. Unpaired vs. Paired
  7. Encoders
  8. Decoder
  9. Embedding Classifiers
  10. Quality Controller
  11. Training and Loss
  12. Pre-training the Auto-Encoder
  13. Pre-training the Embedding Classifiers
  14. Unpaired Training
  15. Paired Training
  16. ...and 14 more sections

Figures (10)

  • Figure 1: Motion retargeting for anonymization of skeleton data.
  • Figure 2: The architecture of DMR
  • Figure 3: The architecture of PMR in the unpaired setting.
  • Figure 4: The architecture of PMR for anonymization in the paired setting. The classifiers are not shown.
  • Figure 5: Example visualization: Actor 40 (Female) performing the "Cross Hands in Front" action.
  • ...and 5 more figures