Human Behavior Recognition Technology Using WiFi-CSI Amplitude And Phase Difference

With the popularization and development of smart homes,human behavior perception technology has received extensive attention and development driven by the demand for smart home applications.Among them,the human behavior perception technology based on WiFi signals is easy to deploy,does not need to be worn,does not infringe user privacy,and It can realize the advantages of through-wall detection,which has attracted the attention of many researchers.The activities of indoor persons will affect the propagation of WiFi signals around.By analyzing the Channel State Information(CSI)of WiFi,and obtaining the influence of the active personnel on WiFi signals,WiFi-based human behavior recognition can be realized.In order to meet the application requirements of human counting and action recognition in smart homes,this paper studies the human behavior sensing technology that combining WiFi-CSI amplitude and phase difference.The main research work is as follows:First,an indoor human behavior sensing technology that combining the amplitude and phase difference of WiFi-CSI information is proposed.In order to simulate the application environment in smart homes,a WiFi-CSI information collection system is built.Two collection schemes were designed for human activities between the transmitting and receiving antennas and human activities on the side of the transmitting and receiving antennas.The CSI Tool open-source software tool installed on the computer realizes the collection of WiFi-CSI information.On this basis,the amplitude and phase difference of WiFi-CSI information are extracted for human behavior recognition,and a new phase difference correction method is proposed for the abnormal CSI phase difference caused by the hardware and software of the acquisition device.The CSI phase difference is processed in segments,which effectively solves the problem of abnormal CSI phase difference jumps.Secondly,a method for human counting and actions recognition based on SVM is proposed.This method uses the sliding window method to extract the time and frequency domain features of the WiFi-CSI amplitude and the time domain features of the phase difference to form a multi-dimensional feature set,and uses the Support Vector Machines(SVM)algorithm to achieve simultaneous detection of the number of people in the room and recognition of actions.The experimental results show that the average accuracy of this method for people recognition in the two experimental schemes are 93.5% and 91.5% respectively,while the average accuracy of action recognition are 97.1% and 94.6% respectively,and it has strong robustness.Finally,a method for human counting and actions recognition based on deep learning is proposed.Aiming at the problems of complex data processing based on traditional machine learning algorithms and the need for human selection of features,a method of using deep learning network to recognize people and actions is proposed.This method uses the amplitude and phase difference information of the WiFi-CSI as the input of the deep network,and uses a two-dimensional convolutional neural network to build a multi-task residual network for recognition of people and actions.At the same time,aiming at the problem mentioned above-mentioned multi-task problem.A multi-task adaptive loss function is designed,which effectively balances the learning ability of the two sub-tasks,and realizes the multi-task perception of the number of people and actions.The experimental results show that the accuracy rates of this method in terms of human counting and action recognition are 100% and 99.4%respectively.Compared with the SVM-based algorithm used in the article,the recognition effect is significantly improved.