Understanding the vulnerability of skeleton-based Human Activity Recognition via black-box attack

Yunfeng Diao, He Wang, Tianjia Shao, Yongliang Yang, Kun Zhou, David C. Hogg, Meng Wang

Abstract: Highlights•We propose the first black-box attack in skeleton-based action recognition. The results show that on-manifold adversarial samples in skeletal motion are truly dangerous because they are not easily identifiable under even strict perceptual studies.•A new adversarial defense for skeleton-based action recognition is proposed, by leveraging the sophisticated distributions of on/off-manifold adversarial samples in adversarial training.•A new perceptual study protocol for evaluating motion attack quality is proposed, addressing that there is currently no metrics suitable for evaluating motion attack quality.