Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Emotion Loss Attacking: Adversarial Attack Perception for Skeleton based on Multi-dimensional Features

Feng Liu, Qing Xu, Qijian Zheng

TL;DR

This work addresses the vulnerability of skeleton-based action recognizers to adversarial perturbations by proposing a multi-dimensional distance that fuses bone-length preservation, bone-angle changes, joint-speed dynamics, and emotion features. An ADMM-based constrained optimization framework is used to generate imperceptible adversarial skeleton sequences, with explicit untargeted and targeted classification losses. Experiments on NTU RGB+D and Kinetics-400 across multiple GC N-based models demonstrate high attack success with lower perturbations than baselines, and indicate that emotion features can modestly enhance imperceptibility and effectiveness. The approach also provides a new metric for measuring skeletal motion distance and offers implications for improving robustness of skeleton-based perception systems.

Abstract

Adversarial attack on skeletal motion is a hot topic. However, existing researches only consider part of dynamic features when measuring distance between skeleton graph sequences, which results in poor imperceptibility. To this end, we propose a novel adversarial attack method to attack action recognizers for skeletal motions. Firstly, our method systematically proposes a dynamic distance function to measure the difference between skeletal motions. Meanwhile, we innovatively introduce emotional features for complementary information. In addition, we use Alternating Direction Method of Multipliers(ADMM) to solve the constrained optimization problem, which generates adversarial samples with better imperceptibility to deceive the classifiers. Experiments show that our method is effective on multiple action classifiers and datasets. When the perturbation magnitude measured by l norms is the same, the dynamic perturbations generated by our method are much lower than that of other methods. What's more, we are the first to prove the effectiveness of emotional features, and provide a new idea for measuring the distance between skeletal motions.

Emotion Loss Attacking: Adversarial Attack Perception for Skeleton based on Multi-dimensional Features

TL;DR

This work addresses the vulnerability of skeleton-based action recognizers to adversarial perturbations by proposing a multi-dimensional distance that fuses bone-length preservation, bone-angle changes, joint-speed dynamics, and emotion features. An ADMM-based constrained optimization framework is used to generate imperceptible adversarial skeleton sequences, with explicit untargeted and targeted classification losses. Experiments on NTU RGB+D and Kinetics-400 across multiple GC N-based models demonstrate high attack success with lower perturbations than baselines, and indicate that emotion features can modestly enhance imperceptibility and effectiveness. The approach also provides a new metric for measuring skeletal motion distance and offers implications for improving robustness of skeleton-based perception systems.

Abstract

Adversarial attack on skeletal motion is a hot topic. However, existing researches only consider part of dynamic features when measuring distance between skeleton graph sequences, which results in poor imperceptibility. To this end, we propose a novel adversarial attack method to attack action recognizers for skeletal motions. Firstly, our method systematically proposes a dynamic distance function to measure the difference between skeletal motions. Meanwhile, we innovatively introduce emotional features for complementary information. In addition, we use Alternating Direction Method of Multipliers(ADMM) to solve the constrained optimization problem, which generates adversarial samples with better imperceptibility to deceive the classifiers. Experiments show that our method is effective on multiple action classifiers and datasets. When the perturbation magnitude measured by l norms is the same, the dynamic perturbations generated by our method are much lower than that of other methods. What's more, we are the first to prove the effectiveness of emotional features, and provide a new idea for measuring the distance between skeletal motions.
Paper Structure (11 sections, 3 equations, 1 figure, 4 tables, 1 algorithm)

This paper contains 11 sections, 3 equations, 1 figure, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Skeleton-based dynamic distance
  4. Classification loss
  5. Emotion loss
  6. Optimal dual method
  7. Experiments
  8. Models and datasets
  9. Evaluation metrics
  10. Attack results
  11. Conclusion

Figures (1)

  • Figure 1: Visual comparison. The green joints represent the original sample and the red ones represent the adversarial sample. (a) shows the original sample, (b) shows attack results of C&W, (c) shows attack results of SMART, (d) shows attack results of our method.