Research On Third-party Source Target Detection Based On Channel State Information

With the development of Internet of Things technology and artificial intelligence,target information in buildings is playing an increasingly important role in the fields of smart home,epidemic prevention and control,and security detection.The existing indoor sensing technology based on the Received Signal Strength Indicator(RSSI)is limited by the unity of the signal strength itself and the need for target cooperation,which affects the accuracy and robustness of its measurement.Therefore,an indoor sensing technology based on Channel State Information(CSI)is proposed in this paper.The ubiquitous WIFI third-party source is used as a passive emission source.At same time,CSI contains both amplitude and phase of the channel,which means it has more fine-grained information,and its sample rate can be controllable on demand.So,it can achieve higher precision and accuracy in complex environments.The main work of this paper is as follows:(1)Aiming at the complexity and diversity of indoor environments,the WIFI channel state information sensing technology and its application in indoor detection are explained and analyzed.The problems of indoor detection are divided into target uncertainty and detection scene uncertainty,and their respective solutions are proposed.(2)In view of the uncertainty of target distribution during detection,a queuing number recognition system based on deep learning is proposed.The system analyzes and compares the impact of various preprocessing techniques on the results.Also,dynamic and static dual models are leveraged to handle different situations that may appear in the queue.Besides,a deep neural network with convolutional long and short-term memory fully connected is used,which effectively overcomes the uncertainty of the target.Experimental results prove that the system can obtain 97%recognition accuracy.(3)In view of the uncertainty of the scene layout during detection,an adaptive behavior extraction technology and two domain-independent recognition models are proposed.The adaptive behavior extraction technology extracts the adaptive threshold of the environment and continuously optimizes the features extracted from the signal to determine the start and end points of the extraction behavior.The core of the two recognition models are the extraction of domain-independent features and the domain adaptation technology.Finally,in gesture recognition and behavior recognition,the feasibility of the two models was verified under the uncertainty of various scenarios such as the personnel location,personnel orientation,and room changes.