Skeleton Based Video Anomaly Detection

Skeleton-based video anomaly detection aims to identify unusual human actions in video by analyzing the movement of skeletal joints, offering a privacy-preserving alternative to appearance-based methods. Current research focuses on developing sophisticated deep learning models, such as those incorporating graph convolutional networks (GCNs), temporal convolutional networks (TCNs), normalizing flows, and diffusion models, to effectively capture spatio-temporal patterns in skeletal data and distinguish normal from anomalous behavior. These advancements improve accuracy and efficiency, particularly through the use of lightweight architectures and innovative methods for modeling motion dynamics. The resulting techniques have significant implications for enhancing safety and security in various applications, including surveillance and healthcare monitoring.