Research On Indoor Positioning Based On Mobile Sensing

With the development of the Internet of Things,the requirements of positioning and navigation in interior scene are increasing.Location information is necessary in different situations,such as path querying in a mall or an airport,goods positioning in a warehouse,pet supervision at home and so on.This thesis studies how to locate with the received signal strength of mobile terminals,based on the structure of "Radio Remote Unit+P-Bridge+Building Baseband Unit".In order to model the relationship between the RSS measurements and the locations,this thesis makes some improvement on a regional RSS fingerprint algorithm based on Back-Propagation neural network,which contains two phases:a training phase and a localization phase.In the training phase,the localization model is trained in each sub-region.In the localization phase,firstly,the user's sub-region is inferred from the characteristics of his/her RSS fingerprint.Then,the trained model is used to infer the relative coordinates of the user.Finally,the absolute coordinates of the user can be calculated according to the sub-region number and the relative coordinates.Simulation results show that the improved algorithm can reduce the positioning error by about 18%compared with the traditional RSS fingerprint localization algorithm.In order to learn from the users'RSS fingerprint sequences collected when the localization model is running,a semi-supervised algorithm is designed in this thesis.Firstly,Baum-Welch algorithm is used to model the movement of the users.Then,the prediction algorithm for hidden Markov model is used to infer the most likely location sequences of the users,and the result is used to label the users'RSS fingerprint sequences.Finally,the labeled samples are used for the incremental training of the neutral network,so that the model can always maintain a high positioning accuracy.Simulation results show that the semi-supervised algorithm can reduce the positioning error compared to only using the measured RSS samples.In order to solve the problem that users lack enthusiasm in RSS fingerprint sensing tasks,this thesis designs an online hybrid incentive mechanism in which users are motivated by reward and virtual credit.The mechanism helps to get the most beneficial RSS data set within a limited budget.Simulation results show that the positioning accuracy of the model is improved more after using the RSS data set obtained by the hybrid incentive mechanism,compared with using the traditional payment incentive mechanism.