Research On The Detection Techniques About Containment Relationship Based On Multiple Readers' Three-state Model

With the continuous development of RFID technology,RFID is applied more and more widely.Containment relationship detection has received extensive attention of academia and industry at home and abroad.Containment exists extensively in real life.Compared with the simple partnership among tag objects,containment relationship is a deeper position relationship.But its process is more complicated.Detecting the specific containment relationship among tag objects can provide high quality service.RFID has a natural advantage of querying the containment relationship for its penetrating characteristics.However,due to the inherent limitations of RFID hardware device,object movement,deployment environment and some other factors,the raw RFID data contains duplicate readings,missing readings and dirty readings.It brings a great challenge to analyze the containment relationship among objects.In order to solve the above problems and infer the containment relationship better,we have done the following work:First of all,considering the problem of low quality and noisy data,we comprehensively consider RFID data characteristics,application restricted conditions and a prior knowledge of the deployment environment.Different from common detecting models,this thesis established the 3-state detection model which has been proved to be optimal to collect the RFID data.Secondly,convert the raw RFID data to the form of source reading matrixes.According to the source reading matrixes,we obtain the probable location sets of objects based on Bayesian Inference.In the process of inferring,we solved the problem of boundary readings' inconformity,which makes the inferring results more reasonable.Then,we will obtain the probable location sets of some tag objects in the monitoring area each time epoch.Construct and maintain a time-varying graph model to represent and statistic possible containment relationship among tag objects according to the probable location sets.Every edge between objects represents their possible containment relationship.each edge will add a statistical vector whose information is to record the positive or negative support of the containment relationship.Thanks to the Bayesian inference of objects' location,the location distribution of each object is represented by the form of probability.This make the quantization of possible containment relationship is more accurate and detailing.Each time epoch update graph model and then infer the containment relationship.In addition,experiments on a large size of simulated data are conducted,which verify the accuracy,efficiency,scalability,flexibility and stability of our method.Finally,we add the module of point-wise mutual information to the above method.This thesis applied point-wise mutual information to the containment relationship detection field.To illustrate clearly,we defines some basic concepts first.Then introduce the mathematical calculation method of point-wise mutual information and its effective implementation in this thesis.Finally,we integrated the point-wise information module into above algorithm.In addition,experiments on a large size of simulated data are conducted,which verify the scalability,flexibility and stability of our method.We analyzed the advantages and disadvantages brought by the point-wise mutual information module and its practical significance.In short,according to the characteristics of RFID equipment and RFID data,we construct the detecting model to clean data,construct time-varying graph model to represent possible containment relationship,using the statics to infer containment relationship and using the point-wise mutual information module to correct the results.Finally,we proposed a complete solution.