A short survey on deep learning for skeleton-based action recognition

Motion recognition is an essential aspect of computer vision used in a wide range of fields and has received much attention as one of the most popular research topics. Traditional motion recognition studies are mainly based on RGB images and videos, but the lighting and viewpoint of RGB data can easily affect the model performance. Skeleton sequences are the most common type of coordinate data and avoid these problems. Therefore, more and more research has been conducted to combine skeleton sequences with deep learning to solve action recognition problems, and awe-inspiring results have been obtained. In particular, the recent rapid emergence of GCN methods, which fit well with the characteristics of skeletal data, offers a promising future for action recognition based on skeletal sequences. In this paper, we first introduce the acquisition of skeletal data and some common datasets, summarise some of the research in the field of skeletal sequence-based action recognition, and briefly discuss the future directions of this kind of research.