The Research Of Human Activity Recognition Algorithm Based On WLAN Channel State Information

The human activity recognition refers to the cognition and recognition of the action types and behavior patterns of the observed objects.As an important research field of Human-Computer Interaction(HCI),human activity recognition has attracted much attention.Traditional vision-based activity recognition technology is susceptible to factors such as light and line-of-sight propagation.Sensor-based activity recognition requires users to wear the special equipment which limit the promotion of activity recognition technology.The human activity recognition technology based on Channel State Information(CSI)of WLAN has aroused concern in recent years because of its low-cost,good concealment and strong environmental adaptability.The existing activity recognition contain the following characteristics: First,the extraction of behavior requires a threshold which is set artificially,and it needs to be reset after the environment changes;Second,most of the researches are only based on characteristics of time domain signal,but ignore the characteristics in frequency domain;Third,the activity recognition algorithm employ a single classifier which has a low recognition rate and poor of robustness.Based on this,this paper presents a human activity recognition algorithm based on CSI of WLAN.The main contents include:Firstly,carry out self-adaption value set algorithm of detection threshold for activity extraction.We focus on using Kernel Density Estimation(K DE)method to obtain the probability density distributions of CSI when there is an action and no action.And nest,find the best detection threshold by comparing their distribution.Finally,combined the time series caching to eliminate the occasional error to complete the activity extraction.Then,the optimal detection thresholds are obtained by comparing them.Finally,the time series caching method is eliminated Incidental error to complete the extraction of behavior.The results show that the algorithm can effectively extract the behavior data in the environment of open and multipath environment.The results from open environment test and multipath environment test show that the algorithm can extract the activity effectively.Secondly,we find the characteristics to distinguishing each activity by analyzing the influence of moving speed on the amplitude of CSI.As for the proble m of information redundancy between CSI subcarriers,the Principal Component Analysis(PCA)algorithm is used to reduce the dimension of CSI,and then extract the time-frequency component information by wavelet transform.In addition,in order to solve the problem that the classification algorithm cannot be used because of the inconsistent length of sequence,we complete the statistical feature extraction to make the feature length consistent.Finally,the features are standardized to prevent the recognition results being affected by some characteristics.Thirdly,the research on activity recognition algorithm is carried out based on feature extraction.According to the characteristics of each kind of activity,a combinatorial classifier is studied by combining the feature subspace and Support Vector Machine(SVM),and then use it to complete the activity recognition.We finish the experiments in both open and multipath environments,the results show that the proposed algorithm can improve the recognition rate of a single classifier effectively.