Research On Equipment-free Target Location And Human Posture Recognition Based On Radio Frequency Signals

Information of localization has become an essential data in many fields,such as military and civilian,which is playing an increasingly important role.Different from the traditional localization methods that most localization systems require that the localization target must carry localization device(such as GPS receiver,mobile phone,etc.)matched with the localization system.However,the target will not carry the corresponding localization device,and the traditional localization method cannot be used at this time in applications such as personnel searching and rescuing,intrusion detection and elderly care under special conditions.In order to solve this kind of localization problem,Device-free Localization(DFL)technology,which does not require the localization target to carry any localization device and actively participate in the localization process,has received extensive attention.Compared with the previous device-free target localization based on camera,ultra-wideband radar,infrared and ultrasonic,DFL technology using wireless sensor networks has become a hotspot in localization research for its low cost,good versatility and the ability to penetrate walls and smoke for localization.At the same time,with the development of DFL technology,relevant techniques which only needs to detect the influence of the target on the electromagnetic field without the active participation of the target is applied to the field outside the positioning,forming a more generalized Device-free Sensing(DFS)study.It is used in the research of gesture recognition,respiratory rate estimation and target speed measurement etc.In this context,this research focuses on "device-free localization" and "gesture recognition" based on the mesured Received Signal Strength(RSS)and Channel State Information(CSI)to achieve more accurate results.The main work of the paper are as follows:(1)The RSS information measurement system based on CC2530 module and the CSI information measurement system based on Intel 5300 network card are built independently.Among them,the RSS information measurement system can realize real-time collection of dynamic information of target in the monitoring area with the characteristics of low complexity,low cost and low power consumption.The CSI information measurement system not only has the functions of the RSS measurement system,but also provides richer time and frequency domain features.(2)In view of the vulnerability of RSS to environmental changes and noise,the background image of Radio Tomographic Imaging(RTI)often inevitably has background noises,and sometimes even the spots of pseudo-targets appearing on the image.In order to improve the quality of RTI imaging,this thesis presents an enhanced RTI method based on sparse self-encoder.This method uses the learning ability of sparse self-encoder to extract the link characteristic information that is effectively affected by the target,and combines with Support Vector Machine performing target extraction to overcome the influence of noise and improve positioning accuracy.The results of indoor and outdoor experiments show that the imaging quality and positioning accuracy of this method are better than the existing RTI method.(3)Based on the Recurrent Neural Network(RNN),the method of human body gesture recognition using RSS and CSI information is presented,and the long-term short-term memory(LSTM)technology is used to enhance the gesture recognition performance of RNN.At the same time,the wavelet transform is used to further extract the features of RSS and CSI information,and then the RNN-LSTM and classical Bayesian classification algorithm are used for gesture recognition.The experimental results show that the proposed method is better than Bayesian classification algorithm.Finally,the research work and the shortcomings in the existing work are summarized.The suggestions are put forward in order to improve in the future research.