Research On Intelligent Perception Technology Of Indoor Personnel Activities In Wi-Fi Environment

With the development of communication technology and Internet of Things applications,the human activity perception technology in the Wi-Fi environment has become the development trend of human-computer interaction in the intelligent age.WiFi signals can realize applications in indoor navigation,smart home appliance control,intrusion detection and other fields by obtaining information such as human body movement trajectory,posture changes,and gesture actions,bringing more convenient experience to people’s lives.However,there are still many problems in this technology,for example,the acquisition equipment is difficult to deploy,and the implementation method is computationally intensive.In response to the above problems,this work uses lightweight data acquisition equipment to realize indoor human activity recognition in a Wi-Fi environment by extracting Channel State Information(CSI).According to different application scenarios and activity types,corresponding perception methods are designed,focusing on the three aspects of indoor gesture recognition,continuous activity recognition and joint recognition of position and gesture for in-depth research.The main research work is as follows:(1)Aiming at the problem of high-precision perception of indoor occupant gestures,an indoor occupant gesture recognition method Wi-Low based on Siamese neural network is proposed.This method collects CSI information through the ESP32 microcontroller,uses a combination of Butterworth filter and discrete wavelet transform(DWT)to remove environmental noise in the original CSI data,and uses principal component analysis(PCA)to reduce data complexity.CNN and Bi-LSTM to extract gesture features,a twin neural network is constructed to realize action classification and recognition.After verification in different experimental environments,the final results show that Wi-Low has an average recognition rate of 93.62% for indoor gestures,and has excellent environmental mobility and robustness.(2)Aiming at the perception problem of continuous motion of indoor personnel,a continuous motion recognition method of indoor personnel Wi-CT under Wi-Fi environment is proposed,and six kinds of daily human behaviors are matched by segmenting continuous motion.Collect continuous motion information of indoor personnel,and use Kalman filtering and PCA to reduce noise and dimensionality of the original data.Segmentation of continuous actions is realized through a dynamic threshold adjustment algorithm,and the Doppler frequency shift correlation value generated during activities is extracted as action features to construct a feature data set.Use Extreme Learning Machine(ELM)to achieve the final classification recognition.Through the verification of the proposed method in three experimental environments,the final experimental results show that the average recognition rate of Wi-CT for continuous motion activities reaches 94.83%,which verifies the feasibility and accuracy of the method.(3)In order to solve the problem of people’s motion perception in different indoor positions,a joint recognition method of indoor people’s position and motion Wi-PA is proposed.Collect 6 types of action information at 16 different positions,and after the original CSI data is denoised and dimensionally reduced,the action segment detection method based on window variance is used to detect the action occurrence domain in the data.One-Dimensional Convolutional Neural Network(1D-CNN)is used to extract the joint features of the person’s position and action information in the sample data,and the joint recognition of the person’s action and position information is realized by building a Residual Network(Res Net).The experiment proves that the average recognition rate of Wi-PA reaches 91.2%,which verifies the feasibility and high robustness of the method.