Research And Implementation Of WiFi Action Segmentation And Real-Time Recognition System Based On Dynamic Threshold

With the rapid development of computer technology,the computing model is changing from being machine-centric to being people-centric.Its future development direction is to make people become a part of the computing link,so as to achieve high-level human-computer interaction,then the recognition and understanding of human actions is an indispensable technical support.Traditional human motion recognition methods include computer vision technology,infrared technology,and special sensor technology.These traditional methods have their own shortcomings.Computer vision is easily affected by light and obstacles.Infrared sensing has a limited range and requires expensive equipment.Special sensors are inconvenient to install and carry and expensive.In recent years,with the popularity of WiFi hotspots and the widespread use of WiFi in the field of perception,human motion recognition technology based on WiFi signals has attracted widespread attention.The channel state information(CSI)of the WiFi signal,as the information of the physical layer,has good multipath resolution capability and can perform fine-grained environment perception.This paper uses CSI to study human motion recognition methods,the main work is as follows:First,this paper designs and implements a real-time indoor motion recognition system based on WiFi signals.This system includes five links: signal acquisition,real-time data transmission,data preprocessing,action interval interception and action recognition,which can classify and recognize indoor single-person actions in real time.In addition,the identification delay and identification accuracy of the system are compared and tested,and the relationship between the two is weighed to obtain a higher overall performance,which proves the effectiveness of the system.Then,this paper proposes a dynamic threshold activity interception algorithm and a multi-window recognition algorithm.One is the action interval interception phase,the CSI amplitude of each action is different,and the optimal threshold is also different.Using the optimal fixed threshold obtained by all types of actions does not change the action interval of each type of action Both are well intercepted,and an active intercept algorithm based on dynamic threshold is proposed for this problem.The second is the action recognition stage.In real life,human actions do not start or end instantaneously.It is shown that each action can intercept multiple action intervals in continuous CSI data.A multi-window recognition algorithm is proposed for this problem.Each action intercepts multiple action intervals,and then classifies and recognizes them separately.All prediction results are weighted using entropy values,and the prediction result with the largest weight is selected as the final recognition result.And use the improved algorithm to redesign and implement the system,and conduct an experimental analysis of the identification delay and identification accuracy of the improved system.The experimental results show that the identification accuracy after the system optimization has been significantly improved,and the identification delay has only slightly increased.Improve the algorithm and improve the effectiveness of the system.