Improved Indoor Positioning System Based On Clustering Algorithm And Feature Extraction

With the development of the Internet of Things and artificial intelligence,the demand for location-based services has gradually increased.Location-based service means that the user sends location information to the server through the mobile terminal,and the service end provides services corresponding to the actual location according to the location information of the user.Due to the rapid development of WiFi technology in the past decade,WiFi have been deployed in most indoor environments.Moreover,the deployment of WiFi hotspots is simple and the price is low.These features are very suitable for indoor positioning.Recently,indoor positioning based on WiFi has has become a focus,and positioning methods based on location fingerprint information have become mainstream.However,the current positioning methods based on location fingerprint information still only remain at the theoretical level.The main reason is that there are usually very large location areas in the scene that require indoor positioning systems.In this scenario,APs are deployed in large numbers and the environment is complex and changeable,which poses great challenges to the positioning system's positioning effect and operating efficiency.This paper aims at this kind of large-scale location scene to carry on the improvement to the traditional fingerprint localization technology,proposed a kind of distributed localization plan based on the clustering algorithm and the characteristic extraction.The main work is as follows:Firstly,this paper researches how to improve the fingerprint dataset with a large amount of data.Each data in the fingerprint dataset corresponds to the RSS value of all APs received by each reference point.In the row direction,we consider using a clustering algorithm to segment the reference point in the fingerprint database and map it to the actual location scene.The area is divided into a small area;after the partition is completed,the RSS values of the APs received by each reference point are not all have a positioning value,and some APs have a low recognition ability for the location.It can not reflect the mapping relationship between the RSS and the location.It needs to extract the features in the RSS information.Therefore,in the processing in the column direction,consider using feature extraction methods to reduce the data dimension.Optimizing from these two directions can reduce the number of data to be positioned and reduce the computational complexity and improve the positioning accuracy.Secondly,select the clustering algorithm and feature extraction algorithm.The K-meansalgorithm and the fuzzy C-means algorithm are the two most widely used clustering algorithms.However,they all have obvious flaws.It is necessary to set the number of categories in advance,and the number of clustering results depends on the number of clusters.The initial cluster center has a very large impact and is highly sensitive to outliers.The use of density peak and distance clustering algorithm(CFSFDP)is proposed and improved by using MapReduce computation framework.In the part of feature extraction algorithm,an improved principal component analysis method based on MapReduce is proposed.Finally,in the online positioning stage,for the K-nearest neighbor algorithm and the weighted K-Nearest Neighbor algorithm,it is necessary to manually set the defect of the K value,and an improved weighted K-nearest neighbor algorithm(EWKNN)is proposed.Compared with the K-nearest neighbor algorithm and the weighted K-nearest neighbor algorithm,this algorithm sets a threshold and dynamically selects the k-value.This avoids the problem that the fixed k value will introduce a point far away from the point where the actual distance deviates from the point to be measured,thereby improving the positioning accuracy.The EWKNN is improved based on MapReduce to achieve parallelization of the algorithm and improve the operating efficiency.Experiments show that the indoor positioning system based on clustering algorithm and feature extraction presented in this paper has greatly improved the positioning effect and operating efficiency compared to the traditional indoor positioning system.