Research On Concurrent Cross-scale And Continuous Activity Recognition Based On Channel State Information

Nowadays,the passive behavior perception utilizing wireless signals has been one of the research hotspots in the field of pervasive computing and real-time systems.It can overcome the high deployment cost,privacy exposure and other problems brought by the behavior perception methods based on sensors or computational vision,and realize the behavior perception method with low cost,low power consumption and non-invasion.This paper focuses on the detection and perception of the concurrency,cross-scale and continuity of human behavior in the natural state,and designs and implements the perception method of concurrent cross-scale behavior and continuity based on channel state information.The main research contents and contributions are as follows.Firstly,a differential dynamic threshold method based on subcarrier correlation is proposed to effectively separate concurrent cross-scale actions from noise-sanitized signals.A method based on multi-layer denoising and variational modal decomposition is designed to extract the characteristics of micro-scale respiratory movement.Secondly,a breath detection algorithm based on power spectral density is proposed.Through judging whether there is breathing in the process of drinking water,the way of drinking water can be determined,so as to make use of the mutual exclusion between swallowing and breathing,look for the deformity of respiratory signal caused by swallowing to complete the swallowing detection,and then estimate the swallowing times in the process of drinking water,and establish the drinking water quantity estimation model.Thirdly,a double-window nested continuous activity segmentation algorithm is proposed,which attempts to use the similarity between actions to identify the continuous activity and estimate the included action sequence.Aiming at the problem of system overhead and recognition accuracy,the action transfer matrix is proposed,and to be combined with activity segmentation to achieve the balance of precision and time cost,and finally the highly efficient perception of non-invasive indoor continuous behavior is realized.Finally,the scheme and algorithm proposed in this paper is evaluated by beingdeployed in real scenarios.The results show that the proposed method outperforms the state-of-art methods in recognition of concurrent activities,segmentation of continuous behaviors,and real-time performance and scalability.