Design And Implementation Of Gesture Recognition System Based On WiFi

With the rapid development of information technology,various smart devices have entered people's daily lives,and human-computer interaction technology has become more and more important.Because of its natural and intuitive,easy to learn,simple and rich information,gestures have become one of the important ways of human-computer interaction technology.The research on gesture recognition technology has become a hot issue.Because of its passive perception,not restricted by lighting conditions,non-line-of-sight,strong scalability,low cost,strong scalability,etc.,the existing gesture recognition technology based on WiFi gesture recognition caused researchers extensive attention.Channel State Information(CSI)is a fine-grained feature of WiFi signals and has high perceptual accuracy.Therefore,using more fine-grained channel state information to realize gesture recognition has gradually become a new trend in the field of gesture recognition at home and abroad..Firstly,a WiFi gesture recognition system based on feature extraction and classification is designed for the current WiFi-based gesture recognition technology,which requires the test subject to face a specific direction and can only recognize the single direction of the gesture.The CSI information of different gestures in four different directions is separately collected and stored,and the amplitude information is extracted as the data set of this paper.Because the directly collected data contains a large amount of noise and the data redundancy is large,the CSI data is first preprocessed such as local anomaly removal and discrete wavelet variation.The mean,standard deviation,correlation coefficient and third-order center-to-center distance of the data are selected as gesture features.The k-Nearest Neighbor(K-NN)and Support Vector Machine Classifier(SVM)are used to train and test the data..Secondly,aiming at the problems of artificially extracted feature types,limited classification of details and cumbersome process in traditional machine learning methods,a WiFi gesture recognition system based on Long Short-Term Memory(LSTM)is designed.In order to avoid data redundancy and ensure the integrity of the gesture information,the optimal subcarriers in each receiving antenna are selected according to the variance size,and preprocessed by discrete wavelet transform.The pre-processed training set data is correspondingly labeled with different gestures,and the LSTM network model is built.The data and tags are input into the network for training and debugging,and the test set data is used for testing.Then the experimental results of the system are analyzed.The results show that the system has obvious effect on multi-directional gesture recognition.The average recognition rate of four gestures is 82.75%,and the recognition rate of ?push? gesture is 87.5%.Finally,the recognition effects of K-NN,SVM and LSTM are compared and analyzed,which shows that the LSTM-based WiFi gesture recognition method has higher recognition accuracy than the traditional machine learning algorithm.The effects of the distance between the transceiver and the data set on the recognition rate and the recognition rate in the line-of-sight and non-line-of-sight environments are analyzed through experiments.