WiFi based human activity recognition with template matching

Wong, Yan Chiew and Saw, Chia Yee and Sarban Singh, Ranjit Singh (2025) WiFi based human activity recognition with template matching. Journal of Engineering Science and Technology (JESTEC), 20 (5). pp. 1403-1413. ISSN 1823-4690

Text 0129810122025181255.pdf Available under License Creative Commons Attribution Non-commercial. Download (1MB)

Abstract

Channel State Information (CSI) based human activity recognition (HAR) has gained popularity because of its privacy-protecting, light-insensitive, and device free benefits. The raw WiFi CSI data obtained from commercial WiFi devices is rich in human activity information and extremely sensitive to changes in the environment, human behavior, and the subject’s weight and height. The data additionally consists of random noise from various sources and interpersonal variations, which may lead to false detection. This research proposes CSI-based HAR with pattern matching that is robust to noise and inter-personal adaptation. Human movements at consistent intervals over a set period of time are converted to pseudocolor images without preprocessing, allowing visual observation of whole pattern changes. The CSI features depicted in the image are then retrieved using a convolution neural network (CNN), and the resulting feature arrays are fed into a stochastic spiking neural network (SSNN) modelbased simple cycle reservoir (SCR) for template matching. The proposed method demonstrates the ability to perform HAR based on partially captured signals, whereby the signal pattern of each segment can be observed in a single line of sight and used for person-to-person template recognition. This enables HAR to address interpatient variability with minimal computational complexity. The experimental results demonstrate that the proposed system achieves impressive performance in recognizing human activities with an overall accuracy of 94.81%. This approach simplifies the original complex CSI data with only 64 features and is robust in recognizing incomplete HAR signals and is capable of forecasting and predicting human activity based on historical data in future.

Item Type:	Article
Uncontrolled Keywords:	Channel state information, Convolution neural network, Human activity recognition, Template matching, Time series prediction.
Divisions:	Faculty Of Electronics And Computer Technology And Engineering
Depositing User:	Norfaradilla Idayu Ab. Ghafar
Date Deposited:	21 Jul 2026 03:21
Last Modified:	21 Jul 2026 03:21
URI:	http://eprints.utem.edu.my/id/eprint/30053
Statistic Details:	View Download Statistic

Actions (login required)

View Item