3D Human Action Recognition: Through the eyes of researchers. (1st May 2022)

• Record Type: Journal Article
• Authors: Sarkar, Arya; Banerjee, Avinandan; Singh, Pawan Kumar; Sarkar, Ram
• Is Part Of: Expert systems with applications. Volume 193(2022)
• Publication Date: 2022-05-01
• Subjects: Human Action Recognition -- Survey -- Datasets -- Gap analysis -- Graph convolutional network (GCN); Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2021.116424
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: Human Action Recognition (HAR) has remained one of the most challenging tasks in computer vision. With the surge in data-driven methodologies, the depth modality has been effectively leveraged to solve this problem by many researchers over the years, mitigating many of the challenges of the two dimensional modalities. In this work, we investigate the methodologies and applications of 3D HAR primarily using the depth modality through a study of the end-to-end process of the state-of-the-art techniques . Challenges, future directions, availability of the datasets, and critical analysis regarding existing methods constitute a part of this paper. This survey aims to provide the reader with an up-to-date analysis in 3D HAR for those interested in further research. Highlights: Focused coverage on 3D HAR including multi-modal, fusion strategies. Discussion of end-to-end pipeline including pre-processing. Critical review of past methods and suggested future directions. Novel Graph Convolutional Network(GCN) based HAR reviewed in detail. Real-world applications of HAR highlighted.