Research On Positioning Fingerprint Indoor Positioning Method Based On Manifold Learning

With the development of ubiquitous computing technology,the demand for location-based services increases day by day and presents a huge commercial prospect.As the key technical foundation,indoor positioning technology has received an extensive attention,and many technology schools have also been derived from this.Among them,the positioning method based on location fingerprint has become an important research direction because it can well inhibit the multipath effect and Non-Line-of-Sight(NLOS)in wireless signal propagation.This paper analyzes the shortcomings of traditional location fingerprint,and conducts in-depth research on the fingerprint database construction and on-line matching algorithm.Firstly,aiming at the problems in the construction of traditional off-line intensive fingerprint database,a set of technical solutions which supports the online update of the fingerprint database is designed.Combining with the large number of signal data and space labeling information generated by the sensor network in the positioning process,high-efficiency reconstruction and dynamic maintenance of the fingerprint database are completed.Specifically,from the perspective of the spatial signal manifold,three key issues in the process of fingerprint location are refined.On this basis,a reconstruction method of position fingerprint based on manifold learning is proposed.By constructing the manifold structure of the spatial signal and combining the corresponding physical coordinate information,the space notation of the signal vector is completed and the corresponding position fingerprint is formed.Through the screening of fingerprint features,the most representative data for the spatial features of the reference point is preserved,and the on-line dynamic update of the fingerprint library is realized.For the matching and positioning stage,a fingerprint matching method combining clustering and classifiers is used for the fingerprint database constructed in this paper.First,according to the spatial distance characteristics of the signal,the initial clustering operation of the original fingerprint database is performed by using k-means clustering,and the clustering results are used to complete the data reorganization of various clusters,thereby limiting the error range of fingerprint matching in the matching process.For the newly formed data subset,the corresponding SVM classifier model is trained and the exact matching for the fingerprint is finally completed.Finally,an indoor positioning system based on low-power Bluetooth is built.The effectiveness of the above methods is analyzed experimentally.Through simulation experiments of fingerprint dynamic iteration process,the proposed manifold learning method can solve the problem of low efficiency of traditional fingerprint collection while ensuring fingerprint accuracy.At the same time,the positioning matching algorithm was verified from two dimensions: classification algorithm and model scheme.Finally,it was shown that the fingerprint matching method based on clustering and SVM could effectively improve the positioning stability and accuracy on the basis of limiting the matching error range.