Research On Crowd Analysis Based On Wireless Spectrum

With the increasing of urban population,the phenomenon of a large number of people gathering is becoming more and more common,and the following group security issues have aroused widespread concern.Since the outbreak of COVID-19,the aggregation activities of people will significantly increase the risk of epidemic transmission,and further promote crowd analysis as a hot topic.Crowd counting and density estimation are important contents in crowd analysis,but there are several problems in the existing research:(1)the visual crowd counting method is affected by crowd occlusion,light conditions and other factors,so it is difficult to obtain satisfactory results in complex environment;(2)The crowd counting method based on WiFi channel state information(CSI)has some problems,such as small monitoring area and serious environmental interference;(3)Due to the influence of media access control(MAC)randomization in mobile terminal,the crowd counting method based on WiFi sniffing is difficult to accurately map the number of MAC addresses to the number of terminals,resulting in the inability to accurately estimate the number of people.Therefore,based on the relationship between wireless spectrum and the number of mobile devices used in the environment,a new crowd counting method is proposed.Firstly,the software defined radio(SDR)is used to collect the I/Q signal in WiFi band,and the short-time Fourier transform(STFT)is used to obtain the frequency-domain information.Then the time-frequency data of I/Q signal is imaged,and the problem of wireless signal recognition is transformed into the problem of image processing in the field of computer vision.Secondly,the convolutional neural networks(CNN),which is widely used in image processing,is used to build a variety of regression models to predict the number of people,namely,LeNet-5＿REG,AlexNet＿REG,VGG16＿REG,VGG19＿REG.The structure of the model is divided into two parts.The front end is the feature extraction part of the classic CNN model,and the back end is the regression prediction layer.Finally,experiments are carried out in real scenes to verify the effectiveness of the proposed method.The experimental results show that VGG16＿REG model has the best performance index,the average absolute percentage error is 13.75%,and the decision coefficient is more than 0.97,which has lower prediction error and higher goodness of fit.Therefore,the proposed crowd counting method has good performance.In summary,this paper carries out the research of crowd analysis based on wireless spectrum for the first time,and verifies the feasibility and effectiveness of the proposed method through experiments,which provides a new research idea for crowd monitoring and analysis.