Research On Passive Human Behavior Recognition Technology In Wireless Environment

Human behavior recognition has been widely used in the fields of identification,humancomputer interaction and so on.In recent years,along with the rapid development of wireless network technology,based on the multipath effects of Wi-Fi signals in indoor transmission,and the human body's occlusion and reflection of wireless signals,a large number of passive human behavior recognition study used Wi-Fi has emerged.Compared with traditional human behavior recognition systems based on wearable devices and video images,the WiFi-based passive human behavior recognition system has the advantages of no need carry any sensor device,easy deployment and non-line of sight.This paper designs and implements a passive human behavior recognition system in indoor environment,and applies it to the two applications of identification and early detection of patients with Parkinson's syndrome.Identity can provide users with personalized services by identifying the identities of different users.By analyzing the walking behavior of the elderly and discovering the abnormalities in the walking posture in time,it can contribute to the early diagnosis of Parkinson's syndrome and improve the quality of life of the elderly.Firstly,the key technical principles of passive human behavior recognition are elaborated,and channel state information is selected as the data source of behavior recognition.The existing feature selection methods and classification methods of Wi-Fi-based passive human behavior recognition are studied,their technical solutions and advantages and disadvantages are summarized and analyzed.Then the selection of Wi-Fi identification information is studied.The effects of human behavior on the amplitude and phase of channel state information in indoor environment are analyzed through experiments.For the current situation that most systems do not use phase information,a choice merge algorithm is proposed.The algorithm introduces phase difference information between the antennas.Aiming at the characteristics of human walking behavior,a preprocessing procedure for channel state information is proposed,including the use of Hamepl filtering to perform abnormal point culing on the original CSI data,filtering the high frequency noise in the signal by wavelet transform,judging the walking behavior by FFT coefficient.It lay the foundation for the application of the two studies on identification and early recognition of patients with Parkinson's syndrome.It proposes a Temporal Convolutional Networks based model to complete the automatic extraction of data features.Compared with the traditional statistical features,the proposed method is more versatile and suitable for the detection of multiple activities without manually develop a large number of time-frequency domain features.Finally,the experimental analysis of the designed system is carried out to verify the performance of the proposed system.The analysis results show that the proposed system can achieve average 97% and 99% accuracy in identification and early detection of Parkinson's syndrome patients.