Research And Implementation Of Human Activity Recognition Based On CSI

With the maturing of pattern recognition theory and the rapid development of Wi-Fi technology,the recognition of human activity based on Channel State Information(Channel State Information,CSI)has been widely concerned by researchers in recent years.The CSI-based human activity recognition method has no need to carry any equipment for its perceptual target,which makes the wireless sensing system easy to deploy and use.It has wide application prospect in secret area security monitoring,home medical care and new human-computer interaction.Although the CSI-based human activity recognition has made great progress in recent years,it still faces many urgent problems.At present,most of the research methods only consider the recognition of human activity,ignoring the characteristics of environmental change,including the location of the human body changes and differences in human activity,which leads to low recognition due to changes in the location of the human body or user differences in practical applications.In addition,the current classification recognition method does not take into account the characteristics of human activity,so how to make the recognition classifier high precision,low complexity,and has a strong generalization ability is a major difficulty in identifying the study.Based on the above problems,this paper mainly carried out three aspects of research work:(1)In the data preprocessing section,the adaptive weighting fusion algorithm is used to fuse the multi-antenna data of CSI,and the local outlier factor(Local Outlier Factor,LOF)algorithm is used to detect the beginning and end of human activity.(2)Aiming at the feature extraction problem in activity recognition process,the relationship between CSI and human behavior in velocity is analyzed,and the wavelet energy and wavelet energy difference on different scales are extracted in the time-frequency domain.With these two features,our system can solve the human body's position changes and the human body's individuality.(3)In terms of the classifier method,we propose a classifier method of FSC-HMM(five semi-connected HMM).Based on the HMM,FSC-HMMdetermines the state number and state-transition conditions that are most suitable for the model according to the characteristics of human behavior,which is more in line with the characteristics of human motion.Therefore,compared with the traditional HMM,it improves the behavior recognition accuracy.Through the research work of the above aspects,the system is designed and implemented,and the verification experiment is carried out.The feasibility and validity of the CSI-based human behavior recognition system are verified by experimental analysis.