Research On Device-free Activity Sensing Via Channel State Information

Indoor human activity recognition remains a hot topic during the last few decades.It has broad applications in smart home,human-computer interaction,intelligent monitoring and other fields.However,the most research approaches require special devices or demand the cooperation of subjects,so the scalability issue remains a great challenge.Later,researchers begin to use our ubiquitous Wi-Fi signal in the passive perception of subject itself or the environment directly,which overcomes the inherent shortcomings of many technical means and has a strong resistance to environmental change,which has less hardware requirements and can achieve a relatively good accuracy.So the passive perception has broad application prospects.This thesis presents a passive and devices-free activity recognition system based on Wi-Fi signals,which can be integrated into any existing WLAN networks.First,we use three algorithms,KNN,Naive Bayes and SVM,to recognize four common activities,namely empty,walking,sitting and standing.By comparison and analysis of many experiments,we summarize some important experimental parameter settings experience.Then we recognize the people's sleeping posture and analyze the results of the identification.Finally,we propose a threshold quality sleep detection algorithm,which can accurately calculate the number of people turning over in the sleeping process.So as to estimate the quality of human sleep,our algorithm has a degree of robustness.In summary,the contributions of this thesis are as follows:First,we present an empirical study on impacts of activities on the Wi-Fi RSSI and CSI,and we elaborate in-depth understanding on the Wi-Fi signal characteristics.Second,we introduce the research that how to recognize indoor human activity through the channel response from theoretical analysis and practical experiments in two parts respectively.We summarize some important experimental parameter settings experience in the actual experiment process.Third,we introduce commonly used sleeping postures through examples,and recognize these postures through some classification algorithms,namely,KNN,NB,SVM,and analyze the accuracy,stability and complexity of the algorithm.We propose a threshold quality sleep detection algorithm,which can detect human sleep quality through calculate the number of turns during sleeping process.