| With the development of computer technology and the expanding application of artificial intelligence,gesture recognition becomes more and more important in human-computer interaction.Duo to the development of wireless communication technology and the widespread use of Wi-Fi devices,gesture recognition based on wireless signals has become a research hotspot.This thesis proposes a gesture recognition method based on channel state information(CSI)and lightweight deep network,aiming at the problems of inconspicuous features and difficulty in extracting high-level features in traditional gesture recognition based on wireless signals.The main research contents are as follows:Firstly,aiming at the problems of low robustness and strong environmental dependence of manually designed features in traditional gesture classification methods based on machine learning,this thesis takes the amplitude and phase information extracted from CSI subcarriers as the research object,constructs RGB pictures containing gesture texture,and automatically extracts high-level gesture features by using deep network.The experimental results show that the method of combining amplitude and phase signals used in this thesis has better performance in classification algorithm than that of single amplitude signals,phase signals and artificial design features.Secondly,a gesture recognition method based on improved lightweight depth network is proposed.A lightweight classification model based on improved MobileNet＿v2 network is used to train and classify RGB images mapped by CSI amplitude and phase matrix.By adjusting the superparameters to reduce the model parameters,the channel attention and spatial attention modules are embedded in the network bottleneck layer to improve the classification accuracy.The effectiveness of the method is verified on a large-scale public data set and a small-scale data set actually collected.Finally,on the basis of single gesture perception,this thesis designs a method of continuous gesture segmentation and detection based on CSI.This scheme is divided into two stages: offline and online.In the offline stage,a large number of CSI data of single sign language movements are collected,and the classification model is obtained by training.In the online stage,STE algorithm is used to segment the gesture segments on the optimal subcarrier,and DTW algorithm is used to detect the similarity of the segmented gesture segments.The effective gesture fragments are input into the classification model trained in the offline stage,and the output classification results are combined to express the meaning of continuous gestures.Through the experiment of sign language sentence interaction,it is proved that this method has a good recognition ability for sign language sentences based on continuous gestures. |
