Recognition Of Human Group Action Based On Wireless Sensing

Today's society is densely-populated and highly complex.With the increasing of emergencies,the monitoring of group become more and more important and difficult.There are many potential applications for group monitoring,such as the statistic of densely populated areas,traffic control in public places and the allocation of service resources.In recent years,with the development of wireless sensing techniques,the Wi-Fi-based method has demonstrated to be a promising method for human behavior sensing and understanding.It doesn't need people carry any devices or provide lighting condition and can achieve the device-free and passive sensing.It has been widely used for human activity recognition,indoor localization and respiration detection.In this paper,we studied how to monitor the group behavior using Wi-Fi devices and the main work can be summarized in following three aspects.(1)We propose a novel method for indoor human detection using Wi-Fi CSI signals.Differing from existing studies that characterize the variation of temporal wireless signals or calculate the deviation of CSIs from a normal profile,our method detects human movements by checking whether there is any phase difference between amplitude waveforms in multiple receiving antennas.We propose the Wi-HD model to estimate the range of sensing coverage and granularity of Wi-Fi based human detection.It is significant for practical deployment and usage of Wi-Fi-based human detection systems(2)We study the human counting method based on Wi-Fi signals and analyze the relationship between the number of human and the variations of Wi-Fi signals.In one hand,the increasing of moving people will cause the increasing of frequency component in signals.On the other hand,it will cause the enhancement of Signal fluctuation.We build the human counting model by characterizing the frequency component of signals and the variations of temporal wireless signals,which is effective in controlled scenarios.Meanwhile,we also analyze the factors affecting the performance of human counting.(3)We propose a crowd density estimation method based on Wi-Fi signals.By deploying the Wi-Fi devices on either side of road and analyzing the signals fluctuations when people move across the line of sight(LOS).Specifically,we design a pedestrian flow detection method using both endpoint detection and sliding window.We extract the effective feature by using principal component analysis(PCA)and time-frequency analysis.By using machine learning method,we can build the model to estimate the crowd density.