Representation for action recognition with motion vector termed as: SDQIO. (February 2023)

• Record Type: Journal Article
• Authors: Islam, M. Shujah; Bakhat, Khush; Iqbal, Mansoor; Khan, Rashid; Ye, ZhongFu; Islam, M. Mattah
• Is Part Of: Expert systems with applications. Volume 212(2023)
• Publication Date: 2023-02
• Subjects: Human action recognition -- Surveillance systems -- Motion image -- Action encoding -- Spatiotemporal features; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.118406
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: We have performed human action recognition termed SDQIO. SDQIO's has been proven to minimize spatial and temporal redundancy in trajectory. SDQIO composed: Steered, diversion, quadrangular, intricacy, obscurity features(s) SDQIO's proposed action descriptor is minimal but successful. Extensive experiments on six standard human action recognition video datasets. Abstract: The active human action motion characteristics from moving areas within a particular spectrum are conveyed by the information extraction domain of video data. Meanwhile, most of the video data frequencies are encoded with silent information that has significant redundancy, resulting in low processing efficiency in conventional video models that accept raw frames as input. Human action recognition remains challenging for the computer vision domain in videos. To address this issue, this study introduces SDQIO, a novel video content extractor that focuses on obtaining multiscale and multidimensional motion information for fast action detection. The ultimate goal of our SDQIO is to develop an effective action recognition module (SDQIO) by identifying five distinct types of action capture operators and extensively assessing their effects on movement prediction over short and long periods. Our SDQIO uses a multi-differentiation modeling approach to gather motion-type information over the whole clip. Extensive experiments revealed that our technique outperforms all other action recognition algorithms by a significant margin on standard action datasets, attaining state-of-the-art and equivalent outcomes.