A Deep Learning Based Lightweight Human Activity Recognition System Using Reconstructed WiFi CSI

Abstract

Human activity recognition (HAR) is a key technology in the field of human–computer interaction. Unlike systems using sensors or special devices, the WiFi channel state information (CSI)-based HAR systems are noncontact and low cost, but they are limited by high computational complexity and poor cross-domain generalization performance. In order to address the above problems, a reconstructed WiFi CSI tensor and deep learning based lightweight HAR system (Wisor-DL) is proposed, which firstly reconstructs WiFi CSI signals with a sparse signal representation algorithm, and a CSI tensor construction and decomposition algorithm. Then, gated temporal convolutional network with residual connections is designed to enhance and fuse the features of the reconstructed WiFi CSI signals. Finally, dendrite network makes the final decision of activity instead of the traditional dense layer. Experimental results show that Wisor-DL is a lightweight HAR system with high recognition accuracy and satisfactory cross-domain generalization ability.