The Research Of Human Activity Recognition Based On CSI

With the rapid development of wireless technology and the universalization of WiFi devices,WiFi signals are widely distributed.Under such circumstances,researchers at home and abroad have begun to study the environmental awareness based on WiFi,in which the human body activity recognition method based on the CSI(Channel State Information)of WiFi physical layer attracts more attention.This dissertation focuses on CSI-based human body activity recognition methods and explores human body activity recognition methods with high accuracy and efficiency.To solve the problem that the existing method does not consider data dimension reduction,cannot effectively eliminate noise and recognize every activity in data containing multiple activities,this paper proposes a fixed path human body activity recognition method WiHAR.WiHAR uses the CSI amplitude preprocessing algorithm proposed in this paper in the preprocessing stage.This algorithm uses the PCA to reduce the dimension and uses the new wavelet threshold function and threshold proposed in this paper denoising,which effectively reduces the computational complexity,improves the efficiency,and can more effectively eliminate the noise.In order to solve the problem that the existing method does not consider the activity interval segmentation resulting from the failure to recognize every activity in the data containing multiple activities.In this paper,the activity interval segmentation algorithm is first proposed and applied to WiHAR.The algorithm uses wavelet function to decompose the denoising results and extract the detail coefficients,then the envelope of the detail coefficients is calculated and the interval of every activity is calculated according to the envelope.WiHAR extracts features for every activity interval and classifies them using a Gaussian Mixture Model to recognize every activity.The experimental results show that WiHAR requires less time to recognize each activity and is more efficiency.WiHAR can accurately recognize the activity of the same person on a fixed path and every activity in data containing multiple activities.To solve the problem that most of the existing methods are not applicable to non-fixed paths,and few of the methods suitable for non-fixed paths have high computational complexity,low efficiency and the reduction of recognition accuracy caused by the difference of activity data,this paper presents a non-fixed path human body activity recognition method NFP-WiHAR.Compared with the existing method using the amplitude and phase of all CSI subcarriers,NFP-WiHAR uses the amplitude and phase of the CSI's nine subcarriers for activity recognition,which not only ensures the accuracy of recognition of non-fixed paths,but also reduces the computational complexity and time required for recognition,and improves the efficiency.In order to solve the problem of declining accuracy caused by differences of activity data,this paper proposes to create an activity template library and the most similar activity extraction algorithm,and apply them to NFP-WiHAR.The most similar activity extraction algorithm calculates the shortest distance between the real activity and the activity in activity template library,and then extracts the activity that are most similar to the real activity.NFP-WiHAR extracts features of the amplitude and phase of 9 sub-carriers for the real activity and the most similar activity,and uses Random Forest for classification.A majority voting mechanism is performed among 18 sub-carriers to select the final classification result,which increases the recognition accuracy of NFP-WiHAR when on non-fixed path and activity data are different.The experimental results show that NFP-WiHAR is more efficient,and can accurately recognize non-fixed path and different people's activities.