Research On RSS Fingerprint Based Low Cost And High Robust Device-Free Localization And Passive Trajectory Depiction

The wireless signal based localization technology,with its relatively simple and universal infrastructure,wide range covering,relatively low price and other advantages,has been at-tracting increasing attentions from both academia and industry in recent years.Specifically,the device-free localization technique without equipping the target with any wireless device and requiring the target to actively participate the localization process,has become a great candidate in emerging applications such as intrusion detection,smart home and intelligent medical,etc.Among different methods,the received signal strength(RSS)fingerprint based method has become mainstream because of its relatively high accuracy and simple opera-tion.In detail,by establishing a radio map(fingerprint database)at the offline stage through recording the distorted RSS measurements caused by the target at each location,and match-ing the real-time measured RSS with the radio map in the online localization stage,the target location can be estimated.Although an important progress the device-free localization technique has made world-wide,existing methods still have the following four main drawbacks in practical applica-tions.First,in the case of slow environmental changes,the fingerprint will change and the radio map will get invalid.The existing methods need to manually re-collecting the RSS fin-gerprints to update the radio map,resulting in a high human effort cost and restricting their practical applications.Second,in the case of sudden environmental changes,the measured RSS at the online stage is affected by the ambient noise and will fluctuate largely.It will lead to the decrease of localization accuracy and limit the system usability in real environ-ment.Third,when the monitoring area changes,the radio maps for different sizes of areas are different.Existing methods need to manually re-establish the radio maps for different areas,resulting in intensive human effort and restricting the practical applications.Fourth,existing passive trajectory depiction methods are based on stitching multiple localization re-sults,it will bring in large computational overhead and high energy consumption,making it inapplicable for real setups.Therefore,it is of great significance to explore the device-free localization and passive trajectory depiction method with low cost and high robustness.Considering the RSS fingerprint based device-free localization,this dissertation analyzes the localization method in the cases of slow environmental changes,sudden changes and area size changes,and makes a goal of low cost,high robustness and low energy consumption.Then three device-free localization methods and a passive target trajectory depiction method are proposed.The main contributions are as follows:(?)For the problem of high human effort cost caused by the manually updating of radio map when the environment changes slowly,we propose a low-cost device-free localiza-tion method iUpdater based on the matrix completion.According to the redundancy and sparseness of the radio map(fingerprint matrix),the problem of fingerprints updating can be modeled as a matrix completion problem on the basis of the regular singular value decompo-sition model.Through collecting fresh RSS of selected reference locations,and combining the unique properties of the fingerprint matrix,the whole matrix can be updated precisely with a low human effort cost.In order to achieve a high localization accuracy,a nonlinear optimization matching model is used.Compared to the existing methods,extensive exper-imental results show that iUpdater reduces the human effort cost at the premise of a high localization accuracy,thus it is more suitable for real applications.(?)For the problem of localization accuracy decreases caused by the online measured RSS fluctuations when the environment changes suddenly,we propose a robust device-free localization method RosenLoc based on the signal decomposition.By decomposing the measured RSS into the target distorted component caused by the target appearance and the noise component caused by the environment noise,we design a Finite State Markov Channel based linear transfer model to extract the target distorted component from the measured RSS.Because the target distorted component is robust to the noise,thus we use this component to establish the radio map.Further,we utilize the dynamic time warping algorithm to match the online target distorted component with the radio map and locate the target.Extensive experiments show that RosenLoc can effectively reduce the environmental influence at the premise of ensuring the localization accuracy,so it has real usability.(?)For the problem of high human effort cost caused by the manually re-establishing of radio map when the area changes,we propose a low cost device-free localization method FitLoc based on the area transfer learning.Since different sizes of areas require to deploy different lengths of links with different RSS distributions,the radio map re-establishment across different sizes of areas can be modeled as an RSS distribution transfer problem.Specifically,we propose a Fisher Linearly Discriminant Analysis based transfer model.By transferring the radio map of original area,it can be shared by the new area and the human cost can be greatly reduced.In addition,the validity of the localization model is analyzed theoretically,and the upper bound of the localization error is proposed.Extensive experi-ments show that comparing to the existing methods,FitLoc can reduce the human cost of radio map re-establishment at the premise of ensuring the localization accuracy,thus it is more applicable for real deployments.(iv)For the problem of high computation cost and energy consumption caused by ex-isting multiple-localization based trajectory depiction method,we propose a low energy consumption passive trajectory depiction method CSTD based on compressive sensing.In view of the time independence and space unity for the locations along the trajectory,together with the sparse property of the location number comparing to the number of all locations,wemodel the trajectory depiction as a compressive sensing based sparse recovery problem.We reduce the amount of data required for trajectory depiction and energy consumption without reducing the depiction accuracy.On the other hand,we propose an Adaptive Matching Pur-suit algorithm to accurately recover the trajectory vector under the condition that the location number along the trajectory is unknown.Extensive experiments show that when comparing to existing method,CSTD can effectively reduce the energy consumption and ensure the accuracy,thus it has a high availability.