Passive Behavioral Awareness And Application Research Based On Device-free Technology

With the rapid development of wireless sensing technology and the rapid growth of human-computer interaction requirements,human activity recognition based on device-free technology has become one of the research hotspots.Compared to traditional computer vision technology and wearable sensing technology,WiFi technology can not only be used to transmit data,but also to sense surrounding environment information in a device-free manner.However,due to the multipath effect and the surrounding environment,the behavioral data based on WiFi signals is highly complex.In addition,the existing human activity recognition methods need to manually extract features,which depends to a large extent on prior knowledge.This thesis takes the elderly living alone as the research object,uses the commercial WiFi device to collect the physical Channel State Information(CSI),and uses the deep learning to realize the research on the coarse-grained and fine-grained human behavior perception.The main research contents of this thesis are as follows:(1)Aiming at human behaviors such as large amplitude and high frequency walking in daily life,this thesis proposes a coarse-grained human activity recognition model LTRNN based on deep learning.The model analyzes the intrinsic relationship between human behavior and CSI data by automatically extracting low-level to high-level behavioral features.In addition,by analyzing the CSI data,the representative subcarriers are selected,and the long-term behavior sequence is segmented into short-term behaviors by sliding the window to improve the efficiency of model construction and training.The experimental results show that the proposed model has better recognition effect under different sliding windows and different sampling rates.(2)Aiming at human behaviors such as small amplitude and low frequency turning over during sleep,this thesis constructs a deep-level fine-grained human sleep activity recognition model BIRM-RNN.The model uses the bidirectional recurrent neural network to learn the context information of CSI data,and inputs the merged context information into the independent residual recurrent neural network to extract the deep data features of the behaviors.At the same time,in order to solve the problem of gradient vanishing of deep network,the model is optimized by Batch Normalization(BN)method.A large number of experimental results in different scenarios show that the model has reached more than 93.5% in each evaluation metric.To sum up,this thesis realizes a passive,no privacy,and no burden method of human activity recognition,and analyzes and recognizes various behaviors of elderly people living alone during daily life and sleep.This method not only makes full use of existing resources,saves manpower and material resources,but also provides new ideas for human activity recognition based on wireless perception,and can be widely used in the field of home health monitoring.