Research Of Indoor Fingerprint Positioning Algorithm Based On Matrix Completion And Dimension Reduction

With the coming of information age,location-based service gradually plays an important role in our daily life.As a powerful supplement of traditional outdoor positioning technology,researches on indoor positioning technology receive extensive attention in recent years.Among many positioning schemes,the method based on location fingerprint has become the mainstream due to its high positioning accuracy and flexibility.In this thesis,the main problems of indoor fingerprint positioning technology are analyzed after a research on principle of fingerprint positioning.Based on Bluetooth technology,the time and space characteristics of signal strength are analyzed by data measured during experiment.Then the thesis is devoted to settle the problems of low efficiency on offline fingerprint building and unfavorable performance during online matching.The detailed work is as follows:In order to improve the efficiency of building fingerprint database on offline stage and reduce working overhead,an algorithm based on non-negative matrix completion is proposed to reconstruct fingerprint.Because the sampling matrix could be accompanied by wild value noise,which can lead to a sharp decrease in positioning accuracy.Thus the wild noise is added to improve the matrix completion model,then he alternating direction method of multipliers is used to get the complete fingerprint matrix.Simulation results show that the method can effectively build the fingerprint,and the reconstruction accuracy is higher than traditional interpolation algorithms.It often contains feature redundancy in the fingerprint database collected in large scenes,which is not beneficial to positioning.And,traditional algorithms searching for all fingerprint database are very time-consuming.Aiming at solving problems above,an algorithm based on semi-supervised affinity propagation clustering and KLDA is proposed.Firstly,the algorithm combines the RSSI of unlabeled point with the offline fingerprint database,then cluster them by semi-supervising.The clustering algorithm was improved by using AP set similarity to reduce singular point when only considering RSSI in clustering process.After completing clustering,KLDA reduction technique is used to remove feature redundancy of fingerprint database.The experimental results show that the algorithm can effectively reduce the location time,and the positioning accuracy is improved obviously when combining with unlabeled data.The mean positioning error reaches 1.86 m.