Research On Radio Tomographic Imaging Based Device-Free Localization Key Technology In RFID Environments

Radio tomographic imaging based device-free localization(RTI)can locate targets without their cooperation and has good application prospects in specific scenarios,such as intrusion monitoring,emergency rescue,etc.It has many advantages of fast speed,easy deployment,and good privacy.Passive UHF RFID tags have many advantages of small size,light weight and low cost,which have been widely used in many industries.Therefore,radio tomographic imaging in passive UHF RFID scenarios has broad application prospects.The indoor environment is complex,the multipath is serious,and the solution of radio tomographic imaging model is an ill-posed problem,which seriously affects RTI positioning performance.To improve its performance,this thesis conducts in-depth research on radio tomographic imaging in passive UHF RFID scenarios.The main works and innovations are as follows:(1)The solution of RTI localization model in passive UHF RFID scenarios is an illposed problem.This thesis analyzes properties of the model and drawbacks of Tikhonov regularization based least square method,and proposes a gradient iterative regularization method(GIRM)to solve the problem.Compared with Tikhonov regularization based least square method that directly gives a regularization parameter that is difficult to determine,GIRM uses a vector of all ones and weight model to derive a regularization parameter required to solve the localization model,and adopts many gradient iterations to further weaken the influence of the vector on solution results.Simulation shows that the accuracy and identification accuracy rate of GIRM are higher than Tikhonov regularization based least square method.(2)When many targets appear in monitored area,wireless signals affected by single target at one moment are hardly estimated.This thesis proposes an unsupervised learning link recovery algorithm and derives a real-time update RTI model(RTURTI)to solve the problem.The measurements at two adjacent moments are compared to obtain signal variations as a single target state changes.The RTURTI localization model is derived by the variations.The thesis researches the law of environmental noise and burst signals,uses the filter and density clustering algorithm to obtain wireless links that are mainly affected by the single target,and then matches it with previously collected single-target database to recover the links simultaneously affected by several targets at two moments.Finally,the RTURTI localization model uses wireless signals of recovered links to locate multiple targets.Test experiment shows that compared with traditional RTI method,the probability that localization error of unsupervised learning link recovery algorithm and RTURTI model is less than 1m is increased by 18% on average.(3)Reconstructed images in RTI contain many false targets and artifact in indoor environments,which affects the judgment of real targets.This thesis proposes a novel method based on gray value cross-sectional scan,gray value distribution analysis,and Naive Bayes classifier to solve the problem.Gray value cross-sectional scan obtains all local maximum gray value pixels in a reconstructed image.Then,gray value distribution analysis extracts characteristic parameters from pixel distributions around these local maximum pixels.Finally,The Naive Bayes classifier classifies all local maximum pixels based on these characteristic parameters,and determines the location and number of real targets.Simulation and test experiment show that gray value analysis and Naive Bayes classifier method can accurately determine the locations and the number of targets from reconstructed images.(4)It is well known that the attenuation of received signal strength(RSS)becomes smaller and smaller as wireless signal propagation distance increases,which decreases the performance of RTI.To improve it,this thesis proposes fine-grained phase parameter based radio tomographic imaging method.The phases of received multi-frequency signals are linearly fitted to obtain phase response(PR),and the offset between target-induced PR and non-target PR is used to obtain phase response shift(PRS).This parameter can reduce the impact of multipath and noise,and does not cause phase ambiguity in RTI.The PRSs at all locations are observed to summarize its weight model,and derive PRS-based radio tomographic imaging localization model(PRSRTI).Test experiment shows that compared with RSS-based RTI,the probability that localization error of PRS-based RTI is less than0.5m has increased by more than 40%,and its identification accuracy rate has increased by more than one time.