Revolutionizing video analytics: a review of action recognition using 3D

As an interdisciplinary field, the 3D video action recognition model interprets human actions in three-dimensional video data using approaches such as pose-based, volumetric, motionbased, and hybrid methods. 3D video action recognition finds application in surveillance, sports analysis, human-computer interaction, healthcare, robotics, and augmented/virtual reality. This paper provides an overview of recent research in 3D video action recognition, concentrating on different deep learning architectures, self-supervised learning, graph-based methods, fewshot and zero-shot learning, cross-modal action understanding, and model interpretability. It also addresses the practicalities of implementing action recognition algorithms in real-world situations, which include tools like deep learning frameworks, pre-trained models, open-source libraries, cloud services, GPU acceleration, and evaluation metrics. Several case studies demonstrate the transformative influence of action recognition in surveillance, humancomputer interaction, sports analysis, industrial automation, healthcare, and retail, with applications in autonomous vehicles, healthcare monitoring, retail analytics, crowd management, video content filtering, and manufacturing quality control. Findings show that integrating 3D video with action recognition algorithms augments accuracy and detail while challenges in video action recognition such as long video features capturing, temporal context maintenance, computational costs management, variability handling, inadequate training data, domain adaptation, and benchmarking methods are addressed. Researchers have contributed novel techniques, architecture, and datasets in an attempt to advance the fiel