Research On Indoor Human Behavior Recognition Algorithm Based On Time-Frequency Joint Analysis

With the continuous advancement of modern technology and life,the need to monitor the activity of human targets in most indoor environments is becoming more urgent.Human behavior recognition in the through-wall scenario has important application prospects in smart home,anti-terrorism and emergency rescue.Due to the widespread popularity of Wi-Fi devices in recent years and researchers' continuous development of Wi-Fi physical layer information,the recognition of human behaviors behind walls by mining Channel State Information(CSI)in the physical layer has received widely attention in recent years.The existing CSI-based human behavior recognition methods through the wall mainly have the following shortcomings: 1.The lack of analysis of the influence of obstacles such as wall when using Wi-Fi equipment to perform human behavior recognition through the wall,and the lack of research on the interference suppression;Second,most of the related researches on behavior recognition use a single feature,and less research on the influence of human activities on the CSI signal itself.Based on this,this paper presents a through-the-wall human behavior recognition algorithm based on Wi-Fi.The main contents include:Firstly,the interference suppression algorithm based on Wi-Fi through-the-wall is studied.In this paper,commercial Wi-Fi devices are utilized to collecting CSI data and the received signal is analyzed and modeled.Then analyze the correlation between CSI amplitude and phase and human activity,analyze and correct the phase error of the CSI signal.And then perform data preprocessing on the signal,including the removal of outliers,detruding and action time series segmentation.Then through discrete wavelet transform to suppress the wall interference and noise interference in the received signal to reconstruct the signal.Finally,through principal component analysis,signal space dimensionality reduction and subspace selection are completed to select signals that are more sensitive to human activities.Secondly,the multi-feature extraction algorithm is studied.First,using Pearson correlation coefficients and Fourier transforms to extract time-domain correlation features and frequency-domain correlation features,respectively.Then extracting the time-frequency features of the signals,and finer-grained time-domain features for similar actions.Thirdly,Constructing a classifier based on the extracted multiple features.A random forest is used to construct a classifier to obtain a robust multi-feature joint recognition classifier.Finally,this paper collects data and performs related experiments in two different environments.Experimental results show that the average recognition accuracy of the system in test environments where the wall material is a hollow wall and a glass wall can reach 92% and 96%,respectively.