Research And Implementation Of Gesture Recognition Method Based On Channel State Information

Gesture recognition is an important part of human-computer interaction,which enables people to interact with the machine naturally and flexibly.This makes the research on gesture recognition attract more and more attention.At present,the research on gesture recognition is mainly based on the following three methods,which are based on computer vision,based on wearable sensors and based on wireless signal.Among them,the method based on computer vision has strict requirements on light conditions,and the target user needs to be exposed in front of the camera,which has the problem of privacy exposure.The method based on wearable sensor requires users to wear hardware devices at all times,and the hardware facilities are expensive,which is not conducive to popularization and applications.Therefore,the Wi Fi-based gesture recognition technology that does not need to consider light conditions or wear any equipment has become a research hotspot.This paper introduces the current solutions for gesture recognition based on wireless sensing technology from three aspects: data preprocessing,gesture feature extraction and gesture recognition.Then,analyzes their problems and propose solutions to this problem to make up for the corresponding defects.This paper proposes a gesture recognition system based on Channel State Information(CSI)called Wi Num,which uses the commercial router to extract CSI in Wi Fi environment and realizes the recognition of ten gestures.The system uses Discrete Wavelet Transform(DWT)method to reduce the noise of the raw data.Compared with the noise reduction method of outlier removal and low-pass filter,DWT can more accurately eliminate the noise information in the gesture signal and retain the CSI data related to the gesture behavior.This paper proposes a novel adaptive gesture segmentation algorithm by constructing the entropy function of the gesture signal.The algorithm can automatically adjust the segmentation threshold according to the current gesture data.And the algorithm can accurately detect the start and end points of each gesture from consecutive gesture signals,which has good robustness and effectiveness.Different from the commonly used gesture classification algorithms,such as K-Nearest Neighbor(KNN),Support Vector Machine(SVM)and Long Short-Term Memory(LSTM),this paper uses Gradient Boosting Decision Tree(GBDT)algorithm for gesture classification.The algorithm can achieve high accuracy under the condition of a small amount of training sample data.In addition,the Wi Num system not only can recognize single gestures,but also recognizes continuous gestures.Finally,in order to evaluate the effectiveness and robustness of the system,this paper conducted an overall evaluation experiment and multiple sets of comparative experiments in the indoor environment.The experimental results show that our system has an average recognition accuracy of 91 % for a single gesture and 85 % for consecutive gestures.The comparison experiment includes the comparison of the different classification methods,the impact of different users and different speeds on the experiment,and the impact of different sampling rates,training sets,and test distance on the system.The results of these six groups of experiments show that the Wi Num system has good robustness and effectiveness.