Research On Gesture Recognition Technology Based On Channel State Information

With the rapid development of computer science and the expanding field of artificial intelligence applications,gesture recognition has become more and more important in HumanComputer Interaction(HCI),which provides technical support for a wider range of emerging applications such as smart home,virtual reality and mobile games.Traditional recognition methods typically rely on a camera or a dedicated sensor worn by the user.Most of the previous researches make improvement in variety and accuracy of gesture recognition.However,there are few related studies on the consistency problem of gesture recognition.Therefore,this paper studies the location-independent and identity-independent gesture consistency recognition based on channel state information(CSI),and tries to enhance the robustness of the gesture recognition system.For the same person's gesture consistency recognition problem at different positions,firstly,in the data preprocessing stage,the dimensionality reduction of the data is processed by principal component analysis method,discrete wavelet threshold transform method is used for denoising and extract gesture actions by variance threshold based on sliding window.Secondly,in the feature extraction stage,four kinds of position-independent features of peak-to-peak,standard deviation,information entropy and average of wavelet coefficient are extracted as feature samples.Next,the feature samples are divided into training sets and test sets,and the training sets and test sets are trained and classified adopting a random forest classifier.Finally,the effectiveness of the proposed method is proved by experiments.For the problem of gesture consistency recognition of different people at the same position,firstly,in the data preprocessing stage,the dimensionality reduction of the data is also processed by principal component analysis method,the Butterworth low-pass filter is selected for denoising and extract gesture actions by variance threshold based on sliding window.Secondly,the feature extraction stage extracts skewness,kurtosis and information entropy which are unrelated to identity and the feature samples of the gesture are normalized into a sequence,which is constructed as a gesture template and stored in the template library.Next,the dynamic time warping algorithm is used to match the gesture to be tested with the template in the gesture template library,and the template with the most similar matching is returned as a result,and the result of the gesture consistency recognition is counted according to the returned result.Finally,the effectiveness of the proposed method is verified by experiments.