Wireless Signal Based Duration Estimation Of Human Motion

With the rapid development of the Internet of Things,human behavior recognition technology is becoming the key to promote the development of medical health,smart home,security monitoring and other fields.Traditional human behavior recognition is mainly realized by cameras and wearable devices.However,the recognition technology based on cameras is limited by user privacy and illumination conditions,and the scheme based on wearable devices is limited by user cooperation.At now,WiFi-based behavior recognition as a new recognition method is attracted the attention of many academia and business circles.This method has the advantages of low cost,no need of light and no user privacy.It can make up for the shortcomings of traditional recognition schemes and achieve better recognition results.There are many wireless channels in the indoor environment where WiFi signal transceivers are arranged.The channel state information(CSI)will change when objectives are moving in the monitoring environment.According to the variance of CSI,the corresponding behavior information can be extracted.Human motion duration is one of the basic parameters in human behavior recognition.The relationship between human activity and disease can be quantified by human motion duration,and the abnormal behavior of moving personnel can be analyzed by it.So far,researchers have focused on specific activity,identification and even breath detection and other aspects,lacking the exploration of human motion duration and ignoring the importance of the basic parameter.Therefore,this paper analyzes the CSI information,and estimates the duration of human motion by combining signal processing and machine learning.The main works of this paper are as follows:(1)The first part realizes the human motion detection by analyzing the change of CSI amplitude.Firstly,the sample data of CSI is preprocessed to remove outliers,high frequency noises,and to complement the missing information.Secondly,the ratio of variance of CSI principal components to the mean first-order difference of eigenvectors is taken as the environmental feature after the dimensionality of the sample data is reduced,and the classification is realized by BP neural network.Finally,the majority voting mechanism is used to make decision on the classification results to judge the current situation of the environment.(2)The second part is to further estimate the duration of human motion on the basis of the first part.Firstly,a CSI sequence with the human motion information is output from the first part and it is divided into several short CSI sequences of equal length,which is based on the idea of transforming the continuous and complex duration estimation of human motion into a discrete and simple problem of human motion detection.Then,the eigenvalues of each short CSI sequence are extracted and the motion detection is carried out using BP neural network model.Similarly,the voting mechanism is used to reduce the systematic errors.Finally,the classification results of each CSI short sequence are synthesized to estimates the human motion duration of each CSI sequence.