Multi-layer Clustering Indoor Location Algorithm Based On Location Fingerprint And Its Application

With the widespread application of fingerprint positioning methods such as Wifi Bluetooth in mobile terminals,indoor high-precision fingerprint clustering algorithm has become a hot research direction in recent years.Reducing the complexity of clustering algorithms and improving the positioning accuracy become technical challenges for fingerprint positioning.Firstly,based on the systematic research on fingerprint location technology,this paper analyzes the indoor wireless signal attenuation model,and conducts indoor wireless signal strength survey,and realizes the fingerprint data transmission in the crowdsourcing process,according to the attenuation model.The correction parameters compensate the fingerprint library to reduce the noise of the fingerprint library.Second,the calculation of the k-means algorithm takes up a large amount of system space and the speed is slow.The SVM neural network algorithm is complex in the process of building the database offline,and the positioning accuracy is not high.This paper proposes the DBSCAN-SVM multi-level clustering algorithm in offline training.The process uses DBSCAN to reduce the number of training samples and quickly find cluster sets that are consistent with the spatial location.The online matching stage uses online cross-validation and grid search SVM algorithms to improve the positioning accuracy.Third,based on the positioning results of the previous chapter clustering algorithm,according to the nonlinear characteristics of the indoor positioning system for the continuity of the dynamic positioning process,this paper proposes an Wifi-DR fusion location algorithm based on extended Kalman filter.Moreover,the similarity of the RSSI vectors before and after the two positioning points is analyzed to generate the system state equation and the observation equation,and the position of the current location point is effectively estimated.In order to support the data resolving of a large number of positioning terminals,and to reduce server processing pressure by 50%.The positioning smoothness is optimized and the continuity of the positioning process is improved with improved positioning accuracy.Fourth,this paper systematically designs and deploys the algorithms and strategies for the first three chapters.A fingerprint acquisition front-end,a server solution,and a database storage are built on the Android platform.The acceptance of the effect of multi-level clustering algorithm for single-point positioning and dynamic positioning scenes was examined.The computational complexity of this positioning system is reduced by about 79.8%in the single point positioning scenario.The positioning accuracy of 2.06m(1 ?)is improved by 56.3%and 37.2%with respect to the K-means and Gaussian mixture models,respectively,and the positioning accuracy is guaranteed while reducing the complexity of the algorithm.In the dynamic scenario,the positioning continuity optimization reduced the processing time variance by 56.6%,and the positioning point variance decreased by 67.4%,which improved the continuity of the dynamic positioning scenario.