| In recent years,with the development of wireless network technology,the method of detecting and recognizing human activities using Wi Fi signals has played an important role in the fields of smart home,medical assistance and emotion tracking,which has been paid much attention and done a large number of researchers.However,most of the existing work focuses on the recognition of single activities.When two activities are both effective,this kind of recognition system is difficult to distinguish the two activities,so it is unable to carry out effective detection and recognition.Therefore,double activities recognition system based on channel state information is proposed in this thesis.The main work of this thesis is as follows:1.According to the theory that the CSI data changes in Wi Fi signal with the signal occlusion,a double activity recognition system based on channel state information is designed to identify the single acitivity category included in the double activity.2.Data augmentation and data preprocessing are performed with amplitude information of CSI signals.The weighted dynamic time wrapping barycentric average algorithm is used to enrich the diversity of samples.Low pass filter is used to remove high frequency noise,and principal component analysis is used to reduce dimension of data to remove redundant information.3.The system uses the Joint Approximate Diagonalization of Eigenmatrices(JADE)algorithm to separate double activities signals into two single-person activity signals,so as to obtain an effective single activity signal.In addition,the system uses the Gate Recurrent Unit network to classify the activity categories.Through automatic information feature learning and sequence information coding,the system can process the sequential CSI activity information and use the softmax layer to predict the corresponding activity categories.4.The system goes with multiple groups of single person and simultaneous double activities in the conference room.The system first goes by single person activity recognition,and the classification and recognition accuracy rate of each type of activity was over 77%.Then,the separation and recognition of double activities were done,and the separation accuracy of three different groups of subjects was 76%,77% and 74%,respectively.Under different experimental conditions,such as standing angle,standing position,activity separation method,activity classification method and iteration number,the separation accuracy of double activities is tested.The experimental results show that the system achieves good robustness and effectiveness. |
