Study On Human Behavior Sensing Technique Based On WiFi Signals

With the rapid development of computer technology and the continuous expansion of the scale of the Internet of Things,the machine-centric computing model constantly shifts towards a human-centric computing model.Realizing a high-level human-computer interaction and enhancing the integration of the physical and information worlds are the essential parts of the development of future computer technology.The existing human behavior sensing technologies,such as computer vision-based sensing,infrared sensing,and dedicated sensor sensing,rely on specific deployment environments and are difficult to satisfy the growing demands of the users on the reliability,security,practicality,and universality.In recent years,with the increasing number of WiFi hotspot deployments and the widespread use of WiFi in the field of indoor location,human behavior sensing techniques based on WiFi signals have attracted wide-spread attention.Compared with the existing behavior sensing methods,the WiFi-based behavior sensing technique has a series of advantages such as it can operate with non-line-of-sight,passive perception(no need to carry a sensor),low cost,easy deployment,no restrictions on lighting conditions,and strong scalability.To further improve the perception and the understanding of the WiFi-based sensing technique,and to promote the continuous development of WiFi sensing technique,this thesis mainly conducts the following research work.(1)Through analyzing the WiFi signal propagation characteristics,we first construct a WiFi signal variation model that reflects the human behavior,and then investigate algorithms to filter the amplitude and phase extracted from WiFi channel state information(CSI)to reduce the noise effect.We design a waveform detection algorithm based on a sliding window and threshold segmentation to separate the action segments related to human daily behavior from the static environment waveform.(2)Optimize the feature extraction method to extract features from the phase and amplitude respectively,and build classifiers based on dynamic time warping(DTW)algorithms and support vector machine(SVM)optimized by the Cuckoo algorithm.Finally,by investigating the multiple decision-making theory,a weighted merge strategy of classification results based on posterior probability and an algorithm to find the optimal weight are designed to realize the daily human body behavior sensing system based on WiFi signals.Experimental simulations show that the classification and merging strategydesigned in this thesis has quite excellent performance on multiple evaluation indicators.