Action capsules: Human skeleton action recognition

Due to the compact and rich high-level representations offered, skeleton-based human action recognition has recently gained more attraction. Although joint relationships investigation in spatial temporal dimensions provides effective information critical recognition, effectively encoding global dependencies of joints during spatio-temporal feature extraction is a prohibitive task. In this paper, we introduce Action Capsule which identifies action-related key by considering latent correlation skeleton sequence. We show that, inference, our end-to-end network pays attention set specific each action, whose encoded features are aggregated recognize action. Additionally, use multiple stages capsules enhances ability classify similar actions. A comparative analysis capsule-based approach with other widely-used methods given, highlighting advantages proposed handling missing data leveraging iterative processing. Consequently, outperforms state-of-the-art approaches on N-UCLA dataset obtains competitive results NTURGBD dataset. This while significantly lower computational requirements based GFLOPs measurements.