Research On Gesture Recognition Based On Channel State Information

With the continuous expansion of the commercial wireless networks and the continuous rise of the smart home industry,gesture recognition and gesture-based stranger recognition have drawn more and more attentions as one of the key technologies.In previous research,people focused more on gesture-recognition systems which based on special devices.Although such gesture recognition systems can achieve higher recognition rates,the application of such gesture recognition systems are not extensive and seriously hindered by the high cost and inconvenient portability of these devices which develop in this area.In summary,the aim of this paper is to propose a kind of gesture recognition method based on the channel state information with high accuracy.Based on the partial techniques used in this method,this paper uses the same gesture of the different people to identify strangers.Aiming at the problem of recognizing many kinds of gestures,this paper proposes a linear transformation algorithm based on phase information to improve the validity of the collected original data in the selection of original data section.Secondly,in the data de-noising section,this paper uses the adaptive threshold wavelet de-noising method to denoise the extracted raw data in the time-frequency domain.Since there is still a small number of sub-carriers that do not conform to the overall trend after de-noising,the denoised data are classified into two categories by selecting some eigenvalues in the time domain and putting them into the classifier to deal with.Thirdly,in the section of extracting abnormal values,this paper proposes an algorithm for extracting outliers based on the variance of sliding window.This algorithm effectively avoids the problem of missing outliers which cased by inaccurate selection of k values in the algorithm of local abnormal factors.Finally,by selecting some eigenvalues in the time-frequency domain and putting them into the KNN classifier,the purpose of accurately recognizing gestures and increasing the types of gesture recognition is achieved.It's different from the solutions proposed in the past,the problem of strangers' gesture recognition is attributed to the recognition of the same kind of gestures of different people.First of all,this paper still uses the phase information which are linearly transformed as the original data.In order to better remove the high-frequency noise mixed in the original signal data and preserve the low-frequency information,this paper uses a Gaussian low-pass filter to denoise the original data.Secondly,in order to more accurately extract the distinctive features of the outliers,this paper proposes a sliding window-based quarter-distance algorithm.By adjusting the test coefficients to improve the algorithm with less coverage defect at the same time extract outliers.Finally,by selecting some features in the time-frequency domain and putting them into the random forest classifier,we can make use of the advantages of the classifier so as to accurately recognize the strangers' gestures.