Skeleton Based Action Recognition

Skeleton-based action recognition focuses on automatically identifying human actions from skeletal data extracted from videos, aiming to create robust and efficient systems for various applications. Current research emphasizes improving model accuracy and efficiency through novel architectures like graph convolutional networks (GCNs) and transformers, often incorporating self-supervised learning and techniques to handle noisy or incomplete data, long-tailed distributions, and out-of-distribution actions. This field is significant for its potential impact on human-computer interaction, healthcare monitoring, and automated analysis of human activities in diverse settings, offering advantages over traditional video-based methods due to its robustness and computational efficiency.