Research On Human Behavior Recognition Based On WiFi-CSI

Nowadays,the wireless network covers a large number of scenes in our lives,subsequently,the gradual progress of wireless sensing technology is continuously affecting our daily lives.In the field of wireless perception technology applied to human behavior recognition,it has important applications in smart home,intrusion detection,medical monitoring,somatosensory games,etc.,in line with the people's pursuit of a smart life,and getting more and more attention from domestic and foreign researchers,and due to the massive deployment of Wi Fi devices,and Wi-Fi-based human behavior recognition technology is more and more favored by researchers due to its advantages such as no need to carry any equipment,low cost,high stability,high accuracy,and high coverage compared to traditional behavior recognition technology.However,the current research on behavior recognition based on Wi-Fi-CSI is also faced with problems such as difficulty in extracting effective features,low perception accuracy of recognition models,and lack of data sets in multiple scenarios.In response to the above problems,this article first adopts a CSI data acquisition system to collect a large amount of channel state information data of human behavior in different environments through a commercial Wi-Fi device that has modified the wireless network card driver firmware to establish a behavior data set.For data classification,this paper establishes a Wi-Fi-CSI-based human behavior recognition models-Wibr.The system mainly consists of data preprocessing,feature extraction and screening,and signal classification.In the data preprocessing stage Use Butterworth low-pass filter to denoise the original signal and reduce the interference of noise on the final recognition effect.In the feature extraction stage,the wavelet packet decomposition algorithm is used to extract the wavelet packet energy ratio feature and the wavelet packet coefficient statistical feature group from the low frequency to the high frequency of the signal.In the feature selection stage,the principal component analysis method is used for the original features with large redundancy.It filters and obtains a feature subset whose dimensions are much smaller than the original feature dimensions but contain most of the original feature information,Improve the system classification rate and reduce the risk of system overfitting.In the signal classification stage,SVM classification algorithm with kernel function parameters optimized by a particle swarm optimization algorithm is used.Through the classification operation on the feature subset,the experimental recognition accuracy is finally obtained.Finally,to verify the robustness and universality of the system proposed in this paper,this article intends to use a large amount of data collected in different scenarios and different sampling rates to verify the effect of the system,finally achieved an average recognition rate of95.6% in different scenarios,it proves that this system has very good feasibility and effectiveness for human behavior recognition.