Hidden Naive Bayes Indoor Fingerprinting Localization Based on Best-Discriminating AP Selection

Indoor fingerprinting localization approaches estimate the location of a mobile object by matching observations of received signal strengths (RSS) from access points (APs) with fingerprint records. In real WLAN environments, there are more and more APs available, with interference between them, which increases the localization difficulty and computational consumption. To cope with this, a novel AP selection method, LocalReliefF-C( a novel method based on ReliefF and correlation coefficient), is proposed. Firstly, on each reference location, the positioning capability of APs is ranked by calculating classification weights. Then, redundant APs are removed via computing the correlations between APs. Finally, the set of best-discriminating APs of each reference location is obtained, which will be used as the input features when the location is estimated. Furthermore, an effective clustering method is adopted to group locations into clusters according to the common subsets of the best-discriminating APs of these locations. In the online stage, firstly, the sequence of RSS observations is collected to calculate the set of the best-discriminating APs on the given location, which is subsequently used to compare with cluster keys in order to determine the target cluster. Then, hidden naive Bayes (HNB) is introduced to estimate the target location, which depicts the real WLAN environment more accurately and takes into account the mutual interaction of the APs. The experiments are conducted in the School of Environmental Science and Spatial Informatics at the China University of Mining and Technology. The results validate the effectiveness of the proposed methods on improving localization accuracy and reducing the computational consumption.