Research On Wi-Fi Gesture Recognition Method Based On Improved TCN

Gesture recognition is an important research direction in the field of human-computer interaction,and has a wide range of applications in smart home,virtual reality and other application scenarios.Compared with methods based on computer vision and sensor devices,the Wi-Fi-based gesture recognition method has the characteristics of low implementation cost and simple deployment,and has certain research value and practical significance.In the task of human gesture classification,it is challenging to achieve efficient and accurate judgment of gesture action in terms of spatiotemporal modeling and understanding complexity.Aiming at single-person gesture action,this thesis studies Wi-Fi signal preprocessing,gesture feature extraction and network model,and combines deep learning technology to build an effective model to improve the recognition effect.Firstly,for the original Wi-Fi signal data containing complex environments,it is difficult to directly extract effective feature information for discrimination.The Channel State Information(CSI)contained in the Wi-Fi signal has the characteristics of strong stability and sensitivity to human behavior.Therefore,the CSI amplitude information is extracted from the Wi-Fi signal as the feature data containing human gestures.as a key carrier for subsequent processing.Secondly,in view of the large amount of environmental interference information in the CSI raw data,this thesis uses the preprocessing methods of Butterworth low-pass filtering,discrete wavelet transform and data normalization to filter and denoise the CSI amplitude data.The input specification extracts slices of data of varying lengths to adapt to the input pattern of the deep learning network.Finally,this thesis proposes multi-input dense Temporal Convolutional Network(TCN)and attention TCN two improved TCN networks to classify and recognize human gestures.Among them,in the multi-input dense TCN,the network calculation amount is effectively reduced by designing a dense connection structure and optimizing the time module,and at the same time,the multi-input structure is introduced for model training;the attention TCN realizes the data by introducing the attention mechanism in the time module and replacing the activation function.Deep feature extraction,and build a progressive structure in TCN to avoid gradient vanishing phenomenon.From the aspects of different networks,sample selection,sub-carrier selection and training time,the two improved TCN deep learning models proposed in this thesis are tested separately.The experimental results show that the average recognition accuracy of the multi-input dense TCN and the attention TCN constructed in this thesis has reached a high level on the three different gesture datasets of Widar3.0,which enhances the original TCN’s ability to perceive and recognize sequence data.It can realize the effective modeling of human gesture recognition in the Wi-Fi environment.