Research On Indoor Localization Algorithm Based On Fingerprint Features

Location-based application services are ubiquitous in people’s daily life,so the research on positioning technology has become more and more important.Currently,the GNSS can basically meet the positioning needs of users in outdoor scenarios and provide high-precision positioning results,but achieving high-precision positioning indoors is still a serious challenge that has not been well solved yet,and more and more researchers are dedicated to obtaining More and more researchers are working to obtain high accuracy positioning results in indoor environments to meet the growing demand for indoor positioning.Fingerprint localization method is one of the most common indoor localization methods,which usually uses RSS or CSI as the localization metric,called fingerprint feature.In the actual indoor localization process,localization systems using a single fingerprint feature often perform poorly because the accuracy of fingerprint localization is affected by the fingerprint type and matching algorithm,and a single fingerprint feature cannot cope well with different localization scenarios.Among them,the traditional RSS-based fingerprint localization method needs to ensure that the dimensionality of fingerprint features in the offline and online phases is the same,but the RSS collected in the online phase is often missing some features than those collected in the offline phase due to environmental changes or human factors and other influences.In addition,the CSI-based fingerprint localization method suffers from the problems of single fingerprint features and incomplete hardware error optimization,which can also lead to poor localization performance.To address the above problems,in order to improve the localization accuracy of indoor localization as well as to reduce the data collection cost,this thesis investigates the localization algorithm for indoor environment,and the related works are as follows:(1)A localization method based on RSS and distributed CSI fingerprint fusion is proposed.Firstly,RSS and CSI data are studied,followed by an improved weighted Euclidean distance algorithm in calculating the similarity of RSS fingerprints,which sets the weights by the relationship between RSS values and location labels.After that,a D-CSI fingerprint generation algorithm is used to obtain a more accurate CSI fingerprint database and reduce the complexity of the computation by dimensioning down the CSI data through the coherent bandwidth principle.Finally,a new confidence calculation method is proposed to fuse RSS and D-CSI features by calculating the obtained confidence,and then obtain the candidate reference points and weight them for target localization.The experimental results show that the fusion algorithm outperforms the localization algorithm with a single fingerprint feature and has certain effectiveness.(2)A migration learning-based indoor fingerprint localization algorithm is proposed.Firstly,the offline database is filtered by the fingerprint features of online samples,and the offline database with high correlation with online samples is retained as the source domain.Then an isomorphic feature space is constructed by learning the mapping relationship between the source domain and the target domain,and the migration matrix is trained in this new feature space to estimate the target location.The experimental results show that the algorithm can maintain high localization accuracy even under the scenarios of environmental changes and the use of different information collection devices.