Research Progress of Skeleton-Based Action Recognition Technologies

Skeletal-based action recognition technology, which analyzes the spatiotemporal sequences of human skeletal joints to identify human behaviors, has garnered widespread attention in computer vision in recent years. This review aims to collate and summarize the research advancements in this domain, with a particular focus on the classification and comparison of feature extraction methodologies. The paper commences by elucidating the acquisition and preprocessing of skeletal data, laying the groundwork for subsequent feature extraction. The thematic focus of the research centers on two predominant feature extraction approaches: those based on customary handcrafted features and those predicated on deep learning methodologies. The methodology encompasses a systematic literature review and comparative analysis, complemented by an introduction to the principal benchmark datasets. The paper juxtaposes the strengths and limitations of various feature extraction techniques through these methodologies and explores their potential in practical applications. In conclusion, this review's importance comes from its thorough analysis of the topic of skeletal-based action recognition. In addition to giving a well-organized summary of the status of the field, it also sheds light on how effective various feature extraction methods are.