Research On Node Optimization For Indoor Passive Localization

As more and more wireless sensors entering into people's life,indoor location-based services have been widely applied into various occasions including health care of the elderly,the fire succor,security monitoring,etc.Among numerous technologies,the passive localization with no devices,which featured unobtrusive perception and non-invasion,can localize and track indoor personnel only based on Radio Frequency(RF),better providing privacy protection.The generation of packet loss and the impact on localization accuracy are caused by the complexity of indoor environment like noise interference,collision conflict and so on.To improve the accuracy,the existing passive localization methods usually deploy overmuch sensor nodes,which may cause the “two-level” redundancy covering node redundancy and link redundancy,leading to high costs in overall localization and low accuracy.Confronting with the above problems,this thesis is focused on the research of the key technologies in node optimization.The main achievements include the following parts:(1)The thesis comes up with a Quick and Robust Localization Approach(QRLA)based on fingerprint to mitigate the impact on localization accuracy caused by packet loss in indoor environment.This method defines three indicators to quantitatively describe packet loss rates in indoor environment.Based on the indicators,the Missing RSS Values Processing(MRVP)is presented to effectively respond to various degrees of packet loss and the Support Vector Machine(SVM)is used to train the classifiers,making high localization accuracy in QRLA.According to the experiment results,QRLA possesses high localization accuracy,packet loss tolerance,etc.Besides,QRLA also features motion insensitivity,which means it can localize the standing and walking samples only by establishing a classifier,effectively reducing the costs in training classifiers.(2)The present passive localization research usually sets the layout and quantity of sensor nodes according to experience.In order to improve localization accuracy,it may cause node redundancy(the first level redundancy)by deploying overmuch sensors,leading to the problems of high costs in hardware and positioning time.A heuristic Redundancy Reduction Method for Indoor Localization(RRIL)method is proposed to solve node redundancy.Based on the genetic algorithm,this method iteratively chooses the subsets of the nodes and indirectly builds the link subsets.What's more,it determines the fitness value of each link subset by building a multi-objective optimal fitness function,creates more chances for the node subsets which have better fitness values to produce the next generation until an approximate optimal solution is found.The research shows that the chromosome coding of the genetic algorithm can easily represent the node combination,effectively reduce the node quantity and shorten the positioning responding time.(3)Analysis reveals that there still exists link redundancy(the second level redundancy)in the node combination selected by the genetic algorithm.This level of redundancy is caused by the consistent changes and strong correlation in link set due to a subject simultaneously disturbing several links within the localization cell.The improved RRIL(RRIL-FE)is proposed based on the feature extraction to work out the link redundancy.When the genetic algorithm generates the subset of localization data,the RRIL-FE can extract a series of orthogonal and uncorrelated quadratic features from the subset by feature extraction.Based on the process,a non-redundant localization dataset will be restructured and used to train a classifier.The study shows that compared with RRIL,RRIL-FE can effectively control node redundancy and link redundancy,reduce node quantity,lower costs in overall deployment and maintain positioning stability.(4)To solve the problem that node quality can not be evaluated by RRIL-FE,the thesis exploratively puts forward the Relief F Group Feature Selection(RGFS)based on the selection of group feature.This algorithm proposes Relief F Group(Relief F-G)emphasized on evaluation of node quality,ensuring the node importance by appraising a related group feature.Based on that,RGFS algorithm is presented on the basis of the sequnential forward selection and wrapper methods.Firstly,this algorithm evaluates the quality of every node by Relief F-G and ranks nodes in descending order;then,respectively selecting nodes from different types,alternative node combinations are built to train classifiers.According to the localization accuracy of each classifier,RGFS decides whether a node can be selected into the optimal node subset or not.It comes to the conclusion that RGFS can acquire the optimal node combination,effectively evaluate the quality of nodes,reduce the time complexity of searching the optimal subset.