Method of sitting posture monitoring based on Wi-Fi channel state information
[Objective]To investigate the sitting posture monitoring method for the problems of existing sitting posture monitoring methods,such as contact,low privacy,high cost,and inconvenient deployment.[Methods]A sitting posture monitoring method based on Wi-Fi channel state information(CSI)is proposed.The method collects Wi-Fi channel state information from commercial routers under different sitting postures,establishes a sitting posture classification model by combining convolutional neural network(CNN)and long and short-term memory(LSTM)neural network,fuses the amplitude and phase data of the channel state information within the sampling window,and fully extracts the spatial and temporal features of the data to improve the sitting posture classification accuracy.When pre-processing the original phase data,a near-neighbor subcarrier difference threshold compensation method is proposed,which effectively solves the problem of phase rotation desynchronization of different subcarriers.[Results]A sitting posture monitoring environment is built to categorize five common sitting postures in office or study scenarios.The experiments prove that the sitting posture monitoring method has high accuracy in classifying sitting postures,and the average accuracy of the classification of all sitting postures reaches 91.23％.Ablation experiments were also conducted to analyze the contribution of amplitude and phase information to the model classification results.The experimental results showed that the model using only amplitude data(average classification accuracy of 91.00％)outperformed the model using only phase data(average classification accuracy of 87.91％).The reason for this is supposed to be that the human body penetrates multiple Fresnel zones of the Wi-Fi signal in different sitting positions.From one Fresnel zone to the adjacent Fresnel zone,the phase value changes.And due to the spinning of phase data,penetrating through multiple Fresnel zones results in periodic changes in phase values,leading to the possibility of the same or similar phase values in different sitting postures.And although the use of phase information alone is not ideal for the classification results,after the model fuses the amplitude and phase data features,the phase information as a reference can play an auxiliary role in the recognition accuracy,and the average classification accuracy reaches 91.23％.[Conclusions]The sitting monitoring method based on Wi-Fi channel state information proposed in this paper has the characteristics of non-contact,high privacy,low cost,easy deployment,and high accuracy of classification of sitting postures,which is of some practical value in the research of sitting monitoring system.The classification model built by combining convolutional neural network and long and short-term memory neural network is conducive to improving the accuracy of the sitting posture classification model.The phase information of the Wi-Fi channel state information also has an auxiliary effect on the sitting posture classification,so it is necessary to do preprocessing of the phase information to participate in the classification of the model in the sitting posture monitoring system.
万方数据
维普
