Non-cooperative Human Motion Detection With Wifi

Human motion detection is an important branch in the application of daily life.Conventional human motion detection contains sensors,cameras and other means.These measures either require the test person to wear additional sensors or be limited by the weather conditions.To tackle these problems,this thesis proposes a method of motion detection with WiFi.Existing WiFi motion detection methods mainly detect motion by utilizing Received Signal Strength Indication(RSSI)and Channel State Information(CSI)of the Wireless Fidelity(WiFi)signal.Since both of them are environment-sensitive,the environment changes will change the values of RSSI and CSI.In many designs,the former is more developed and the most typical prototype is to use RSSI fingerprint database to determine the location of human body and then determine the trajectory of the human body.The disadvantage of this method is that RSSI is not enough in accuracy.The use of CSI solves this problem,but how to extract motion-compensated CSI and map CSI to human motion is challenges.In addition,the body motion includes body movement and movement of human limbs.Different movements may cause different range of movement,thus the changes of the environment is also different.How to deal with different range of motions by using the corresponding solution is also a difficult task.To cope with the two difficulties above,this article proposes two solutions to solve the problem of how to improve the accuracy of the two different motion detection such as body motion and gesture.For body movement(human walking),considering the cost factor,this paper uses wireless network card as a CSI collection equipment.Due to the low accuracy of the wireless network card,this paper proposes the use of linear regression to correct the error caused by the local clock is not synchronized with WiFi signal.Considering the changes of CSI caused by walking is larger,the Local Outlier Factor(LOF)and Hampel filters are used to correct abnormalities.Considering the computational efficiency,we use only the largest eigenvalue of amplitude and the largest eigenvalue of phase as the motion characteristics to map to walking or rest,and use a simple classification algorithm to obtain high accuracy.Aiming at the finer human motion of gesture movement,a more sophisticated equipment Universal Software Radio Peripheral(USRP)is used as a CSI collecting device,a more accurate signal analyzing method wavelet transform is used to remove CSI data noise to extract the CSI representing the gesture,and some more complex classification algorithms(Random Forest and Convolution Neural Network algorithm)are used.Considering the discontinuity of WiFi packets transmitted by wireless routers,this paper proposes a method of time synchronization by using timestamps so that cutting CSI data blocks into pieces is available.Finally,the method proposed in this paper can detect the body walking with an accuracy of 95% and the four gestures with an accuracy of 92% correctly and quickly.