Research On Aperiodic Activity Sequence Recognition Based On WiFi Channel State Information

As the core technology of intelligent life,activity perception can greatly improve the quality of human-computer interaction and expand the range of intelligent applications in fields such as daily activity recognition,gesture recognition,vital signs monitoring,crowd dynamic tracking,location,identification and other aspects.With activity perception playing an increasingly important role in human daily life,more complex and changeable application scenarios ask for higher technical requirements for human activity perception.However,most of the existing research on daily activity recognition based on WiFi Channel State Information(CSI)still can not identify the large and aperiodic activity sequences in real time.In order to solve this kind of activity sequence recognition problem,the paper proposes a novel kind of activity sequence recognition scheme by a combined deep learning network based on short-term Principal Component Analysis(SPCA)to reduce data redundancy and beam search to optimize recognition accuracy,which can recognize aperiodic activity sequences with fuzzy boundaries and bad consistency conditions under terrible conducting.The main research results of the paper are as followed:(1)The paper proposes an activity sequence recognition scheme based on WiFi CSI.Firstly,the WiFi-based perception model proposes to use CSI amplitude information to extract the features and realize the largescale aperiodic activity sequence recognition.It is proposed to extract local individual activity features by CNN network and extract sequence features by LSTM network,establishing a combined network for the recognition model of large-scale aperiodic activity sequences.(2)The paper proposes a real-time activity sequence recognition scheme,in which the preprocessing module uses real-time interpolation and filter methods to reduce noise,as well as the SPCA with sliding window to reduce redundancy.Then the CSI streams are segmented into clips to match the SPCA sliding window.The combination network model is followed to work.(3)The paper uses experiments to decide the deployment and configuration of antennas.After the analysis of Fresnel zone,the first Fresnel zone is used as the activity conducting zone for it has the strongest ability to sense the changed WiFi signal.Based on the extracted effective CSI,the CNN network extracts activity features that are more related to the activity itself.(4)The paper proposes to use the beam search method to optimize the recognition accuracy of the combined network.The paper changes a little bit of the basic beam search method to make it suitable for the scene and get around the fuzzy segmentation problems without initial states between two activities.It reduces the possibility of inaccurate marks and improve the recognition performance.First of all,the paper introduces the background and significance of activity sequence recognition research,in addition to analyze the current research status and shortcomings of various sensing technologies.Then it proposes the technical route of the research after introduceing relevant technologies.On this basis,the technical details of the WiFi-based activity recognition scheme are presented in detail.Finally,the performance of the scheme is evaluated through a large number of experiments with the yogastyle activities,which proves that the paper is of great significance in the application of generally large-scale aperiodic activity sequence recognition.