Title: AROS: human action recognition by spatio-temporal fusion mechanism based on optimised subcarriers

Authors: Zhiyong Tao; Xijun Guo; Ying Liu

Addresses: School of Electronic and Information Engineering, Liaoning Technical University, Huludao, 125000, China ' School of Electronic and Information Engineering, Liaoning Technical University, Huludao, 125000, China ' School of Electronic and Information Engineering, Liaoning Technical University, Huludao, 125000, China

Abstract: Wi-Fi-based human motion recognition methods are widely used for their usage and infrastructure convenience. However, the spatial diversity of MIMO causes differences in the representation of action features across antenna links. Furthermore, using all of the data is computationally time-consuming, and using a portion of the data may result in omitting crucial features. To address these issues, a human action recognition by spatio-temporal fusion mechanism based on optimised subcarriers (AROS) is proposed in this paper. Specifically, K-means adaptive clustering aims to select information-rich and complementary subcarriers with adaptive clustering cores through cluster analysis and correlation computation. The correlation-weighted fusion mechanism is presented to enhance the MIMO link characteristics. A network structure based on spatial module, temporal convolutional network, and temporal attention is presented to extract CSI spatio-temporal features. Experimental results demonstrate the effectiveness of AROS, achieving accuracies of 96.17%, 96.31%, and 94.51% in three different environments.

Keywords: human activity recognition; HAR; channel state information; CSI; temporal convolutional network; TCN; subcarrier selection.

DOI: 10.1504/IJSNET.2024.140389

International Journal of Sensor Networks, 2024 Vol.45 No.4, pp.204 - 215

Received: 15 Dec 2023
Accepted: 07 May 2024
Published online: 06 Aug 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article