Research On Indoor Localization Based On Recurrent Neural Network

The indoor localization method is one of the hot research directions in the current localization method.As the Internet of Everything gradually enters people's daily life,the realization of location-based services has become one of the most urgent needs today.Due to the complex of the indoor environment,the fingerprint localization method has superior performance in indoor positioning.However,the current fingerprint localization technology still has many problems and challenges to be solved.Most fingerprint localization methods are based on single point matching.The traditional fingerprint database needs to measure the fingerprint information and mark the position at each reference point,which takes a long time to collect the fingerprint data.In order to reduce labor costs,researchers have proposed fingerprint information collection methods such as crowdsourcing or automatic robot acquisition.The data collected by these methods are often on one path.In order to be able to localization by using current traditional methods,it is often necessary to reconvert these path sequence data into grid point data.Based on the current researches of fingerprint localization at home and abroad,this thesis focuses on using the recurrent neural network to realize the target indoor localization.This study aims to solve the existing problems more or less and includes the following aspects:Firstly,for the fingerprint data,this thesis realizes the indoor environment model by simulation method and provides a simulation environment to valid localization performance for the latter localization methods.This thesis uses ray tracing method to restore the propagation of electromagnetic waves in indoor environments.The signal is abstracted into a large number of rays which radiate into free space,then each ray is tracked and the signal components are collected at the receiver to model the indoor environment.Secondly,according to the limitation of the indoor environment structure in the indoor environment,the user only can move in a limited set of trajectories.This thesis proposes an indoor localization algorithm based on double layer recurrent neural network.The method uses the constraints of historical information to estimate the current location.The experimental results in the simulation environment and the real world environment show that the proposed algorithm has better localization performance and robustness than traditional fingerprint localization algorithm.Finally,this thesis proposes an indoor localization algorithm based on the encoder-decoder model.The network structure of the algorithm is simpler than the double layer recurrent neural network model.The experimental results of the simulation environment and the real world environment show that the localization performance of the method is similar to the previous method,and the training time is shorter.The method can effectively reduce the training time under the condition of a large amount of data which has important significance for the application of indoor localization.