Human Movement Recognition using Deep Learning on Visualized CSI Wi-Fi data

Abstract

Wireless signal transmission is an intricate process, significantly influenced by the environment within which it operates. Notably, the mobility of various elements within this environment, such as the parts of a human body, distinctly modifies the manner in which these signals are reflected. These alterations subsequently cause changes in Channel State Information (CSI) data captured by Wi-Fi routers. Intriguingly, specific human behaviors can be detected through a meticulous examination of the data streams from CSI. These behaviors, representing diverse activities, can be identified by processing the data streams and juxtaposing them against predefined models. The recognition of these activities hinges on discerning patterns within the CSI data, reflecting the relationship between human movement and the variation in channel state information. A variety of techniques have been developed to explore and understand these patterns, with machine learning emerging as the most popular and effective tool. Machine learning techniques are harnessed to develop sophisticated models capable of correlating variations in channel state information with specific human movements. These correlations enable the prediction and identification of human activities based on changes in CSI data. This research focuses on further exploring this intriguing intersection of human activity, wireless signal processing, and machine learning. It aims to provide a deeper understanding of these correlations and develop more effective models for human activity recognition. More specifically, with this work we attempt to to explore new way of using the CSI data in Deep Learning tasks. That is by using the visualized amplitude of signals and correlate them to certain activities.