Research On Multi-scene Personnel Detection System And Algorithm Under Wi-Fi

Indoor human perception technology has a wide range of applications in the fields of personnel detection,indoor positioning,motion analysis and security detection.Due to the universality and low cost of passive detection,the use of commercial Wireless Fidelity(Wi-Fi)for personnel detection has gradually become a popular research problem.The traditional human body sensing technology generally adopts a method based on Received Signal Strength Indication(RSSI).This method has the problems of poor stability and low positioning accuracy due to strong multipath interference and time-varying signal.Due to its high accuracy and robustness,the perception signal of Channel State Information(CSI)has gradually become the base signal for human perception.main tasks as follows:(1)Developed a CSI collection and analysis system based on commercial Wi-Fi equipment.This system can set CSI signal collection based on two different network cards based on Atheros and Intel 5300 according to different network cards.It can also adjust the working frequency band to work under 5G and 2.4G,and it can replay and denoise signals.And dimensionality reduction,time and frequency domain analysis of a single signal,and machine detection algorithms can also be used for human detection.(2)An Radial Basis Function K-NearestNeighbor(RBF-KNN)personnel detection algorithm suitable for simple environments is proposed.The algorithm uses Kalman filtering to denoise the signal during the offline phase and uses the radial basis Function(RBF)based on the online phase.Function improved KNN algorithm for matching method.After experiments,the algorithm has high recognition accuracy and fast recognition speed in a simple environment.(3)In order to solve the problem of human detection in a complex environment,this paper proposes a Support Vector Machine-Channel State Information(SVM-CSI)algorithm.This algorithm reduces the CSI through Principal Components Analysis(PCA).The amount of signal data is reduced,the main features of the signal are extracted,and then a regression model is trained by Support Vector Machine(SVM),and the final result is obtained according to the correspondence between CSI signals and actions.The algorithm passes The detection error of the experiment in simple environment and complex environment is not more than 5%,it has strong robustness,and it can reach 91% recognition accuracy in complex environment.