Research On Behavior-aware Techniques Based On Channel State Information

With the concept of "Internet of things" proposed and the wide deployment of indoor wireless access devices,Wi Fi-based environmental perception technology can realize the perception of the surrounding environment through the existing wireless devices.At the same time,it has very important research significances in many fields,such as personnel action detection,medical monitoring,indoor positioning and so on.The technology is mainly based on the influence of the movement of the objects in the environment on the transmission link of wireless channel,collecting two major characteristics of wireless channel: Received Signal Strength Indicator or Channel State Information,and studying the feature of the collected information,so as to realize the mapping from wireless signals to human behavior.Among them,RSSI-based sensing technology is vulnerable to multipath interference,which leads to the final recognition accuracy is not high.Compared with RSSI,CSI from the physical layer of wireless communication can describe the channel state of Wi Fi signal in detail from the scale of subcarrier,which can describe actions more finely and achieve high precision perception.In this thesis,several typical environmental perception technologies are introduced in detail in combination with existing work,and the thesis analyzes and summarizes the application schemes,advantages and disadvantages of each technology.At the same time,the behavior sensing method based on Wi Fi is deeply studied,the thesis provides solutions to two hot issues in this research direction,namely,the CSI-based rehabilitation action recognition system and the CSI-based suspicious object concealed by pedestrian detection system.In the CSI-based rehabilitation action recognition system,the system uses CSI waveform with both time domain and frequency domain information as action feature,and the action signals are preprocessed by Hampel filter,Butterworth filter,Linear Discriminant Analysis algorithm and other signal processing techniques.Two kinds of data augmentation techniques are designed to expand the experimental dataset in the system,realizing low-cost and high-precision rehabilitation action recognition.On this basis,the system then scores the rehabilitation degree with the help of the authoritative functional evaluation standard.In the CSI-based suspicious object concealed by pedestrian detection system,to reduce data complexity,system uses Principal Component Analysis algorithm to reduce the dimension of highdimensional signals,and the system uses Convolutional Neural Network to classify different categories of objects,so as to realize the detection of suspicious object carried by pedestrians in indoor environment.At last,a large number of experiments are carried out to evaluate the performance of the two systems.The result shows that systems based on CSI are feasible and robust.