Research On Precision Improvement Technology Of Wi-Fi-Based Behavior Sensing System

Behavior sensing is a method for machines to analyze and understand the behavioral characteristics of objects such as humans or animals.Behavior sensing technology based on Wi-Fi Channel State Information(CSI)has gained widespread attention because of its advantages of low cost and wide coverage.However,Wi-Fi-based behavior sensing is affected by the channel bandwidth,environmental conditions and other factors,and has the problem of low sensing accuracy.In this paper,we address the limitations of commodity Wi-Fi devices in behavior sensing,and investigate aggregated bandwidth and multi-device collaboration as examples,aiming to improve the accuracy of Wi-Fi-based behavior sensing systems,respectively.To solve the problem of insufficient sensing resolution due to the limited bandwidth of Wi-Fi channel,this paper proposes a mechanism to improve the sensing resolution of behavior based on Wi-Fi.The CSI is collected from several adjacent Wi-Fi channels in a very short time interval by the fast switching mechanism of multiple Wi-Fi channels,and the CSI of the aggregated channel is obtained by the aggregation mechanism of adjacent Wi-Fi channels,and the bandwidth of the Wi-Fi channel is extended to the full bandwidth of the switchable channel.The feature extraction module of "time-frequency-spatial domain" is constructed to complete the feature extraction of aggregated CSI.The experimental results show that the accuracy of behavior recognition is improved from 88.6%to 96.4%by using the proposed resolution improvement mechanism,which effectively improves the sensing resolution.To address the problem of changing environmental conditions affecting sensing precision,this paper proposes an optimal sensing link dynamic intelligent selection mechanism for multi-Wi-Fi device collaboration scenarios.Formulating Wi-Fi link selection as a sequential decision problem,a deep reinforcement learning agent is constructed,and reinforcement learning is used to solve the problem.Multiple links are intelligently weighted and ranked according to the dynamics of the Wi-Fi environment and objects,and the link that best characterizes situational information such as target pose,activity and location in the region is selected by the agent.The relationship model between the wireless signal and the state of the target in the corresponding region is constructed based on the CSI of this link.The experimental results demonstrate that the mechanism can achieve the approximate performance of multiple links with the collaboration of multiple Wi-Fi devices using only one Wi-Fi link.