Research And Implementation Of WIFI Gesture Recognition Based On Domain Adaptive

With the continuous development of Internet of Things technology,human-computer interaction technology applied in smart home and medical health scenarios has become more and more important.Gestures have become an important method of human-computer interaction because of their convenience,easy-to-understand,and rich meaning.At the same time,wireless perception technology has also been continuously broken.WIFI devices are widely deployed in living environments.Therefore,the research on WIFI-based gesture recognition has become the hot spot.In addition,compared with video and device-based gesture recognition,the use of WIFI for gesture recognition has the advantages of not being affected by light and no need to wear special equipment.This thesis is a research on indoor gesture recognition of Channels State Information(CSI)of commercial WIFI systems.WIFI signal is very susceptible to environmental factors,and the general gesture recognition method is to recognize after collecting data in a single environment.If the gesture recognition model in one environment is directly applied to another environment,the accuracy will be very low.Aiming at the problem that the existing WIFI-based gesture recognition methods can only accurately recognize gestures in a single environment,this thesis mainly designs an adaptive gesture recognition method in a field independent of the environment.Only the labeled data in one environment and the unlabeled data in another environment are used to predict the label of the unlabeled data.This approach greatly reduces the cost of data collection.Based on this,the research work of this thesis is as follows:(1)Aiming at the problem that each gesture in the gesture data in the current gesture recognition method is relatively single,each gesture collected in the laboratory and conference room in this article is more diverse.After collecting gesture WIFI information,first extract CSI amplitude information,and then use discrete wavelet transform to denoise.In order to intercept the complete gesture segment,a buffer-based gesture segment interception algorithm is proposed for the data collected in this thesis.Experimental results show that this algorithm is better than the existing gesture segment interception algorithm.(2)A CNN-Bi LSTM deep learning model based on the attention mechanism is proposed for WiFi gesture recognition.Aiming at the characteristics of the WIFI CSI gesture data set,the CNN model is introduced to extract the local features of the gestures,and Bi LSTM is also used to consider the timing features of the gestures.Experimental results show that the average recognition rate of this method in the four gesture categories is increased by 3.4% compared with the existing advanced algorithms.(3)Based on the domain adaptive framework ADDA,an environmental adaptive gesture recognition method suitable for WIFI CSI gesture recognition is implemented.At the same time,CNN-Bi LSTM is used to extract gesture features in laboratories and conference rooms.The results show that the use of this model in WIFI CSI is 2% higher than the other domain adaptation models.