Indoor positioning using Wi-Fi fingerprint with signal clustering

Abstract

Indoor positioning systems have been actively studied so far, and recently, many researchers are adopting Wi-Fi signals for their systems. It is mainly because Wi-Fi networks are prevalent in the indoor environments these days; also it is more efficient than other methods as one can estimate his location by simply comparing current RSS (Received Signal Strength) with the fingerprint of Wi-Fi signals measured in advance at the area. One of the simple and popular approaches for matching an RSS with the Wi-Fi fingerprint, is known as K-Nearest Neighbors (KNN). While KNN uses all data stored in the fingerprint for the comparison, it shows relatively low accuracy due to the unstable nature of Wi-Fi signals: Especially in a spacious or congested place, its performance may severely degrade. In this paper, we adopt K-means clustering for an efficient and accurate classification in KNN. In our approach, in order to mitigate the effect of unstable Wi-Fi signals, the clustering is performed using every individual RSS values rather than representative mean values, which results in the elimination of extreme RSS values in the classification. In addition, locating the position by KNN utilizes the probability distribution of RSS values in each cluster. Under a wide range of K in KNN and the number of clusters, our experiments show notably enhanced performances than those of the classical KNN.