Research On Improved Indoor Positioning Algorithm Based On RSS

It is difficult to maintain high positioning accuracy in indoor edge areas for the existing indoor positioning technologies,such as Bluetooth,UWB,Infrared,Geomagnetic,Zig Bee,etc.,Multiplepoint localization is one of the most common methods of indoor positioning,which has a simple principle and is easy to implement in practice.However,estimating distance through path loss model will produce great errors.Fingerprint based localization methods possesses a high accuracy,nevertheless,the timeliness of fingerprint databases brings serious challenges in practical applications.Focusing on the bigger point location error in the edge area,we provide a positioning error rectification method based on K-means clustering and machine learning.Furthermore,we propose self-adaption path loss model selection method to improve positioning accuracy and reduce the cost of data acquisition.The main contributions include:1.In this thesis,a positioning error rectification method is proposed to improve the positioning accuracy of the indoor edge area.This method sets up the corrective benchmarking points by multiplepoint localization,boundary judgment,and K-means clustering.In the online stage,the RSS data of the target node is obtained,and then the multiple-point localization is employed to estimate the target node position.Then,the positioning result is corrected by corrective benchmarking points.Finally,the best matching rate was improved through double space determination and convolutional neural network.The simulation results indicate that,by the proposed method,the positioning accuracy of the indoor edge area is effectively improved.2.An adaptive positioning algorithm based on multi-path loss model is proposed to improve the errors produced by estimating distance.The collected RSS data is used to fit multiple path loss models by this algorithm,then the optimal combination of path loss models is determined by the iterative least square method and error function.Finally,the positioning result is optimized by the optimal combination of path loss models.In addition,this thesis also proposes a multi-path loss model algorithm based on federated learning,which avoid direct data uploading through model interaction,thus user privacy can be protected while ensuring positioning accuracy.The simulation results indicate that,compared with the two-sloping model,the positioning accuracy of this algorithm mentioned in this thesis is higher and more stable.Besides,it is validated that the algorithm based on federated learning also achieves approximate performance of the original algorithm.Based on the above two works,the positioning accuracy of the indoor edge areas can be improved by setting up the corrective benchmark points and multi-path loss models with the aid of the first positioning results.