A New Set of Bluetooth-Based Fingerprinting Algorithms for Indoor Location Services

Indoor Location Based Services (LBSs) are likely to make significant contributions to society and the economy in the near future. Unlike outdoor LBSs, the technologies and methodologies used are still very much under development, and there are a number of challenges which must be addressed, namely: accuracy, precision, and time complexity. For the accuracy metric, the Euclidean distance error is calculated based on the difference between the known location (points of interests) and the localized position. Regarding the precision metric, the distribution of the distance errors is computed. In this paper, new fingerprinting-based algorithms, namely, the Nearest Neighbor Version 2 (NNV2), Nearest Neighbor Version 3 (NNV3) and Nearest Neighbor Version 4 (NNV4), are proposed, and tested to determine the most effective and efficient one with respect to those challenges. Our analysis reveals that: (1) the time complexity for each of the Nearest Neighbour (NN) and KNN algorithms (i.e., K is constant) is $(1\ast \mathbf{n}\ast \mathbf{m}+1\ast \mathbf{m})$ -comparison which is more than that for NNV2 and NNV4 (i.e., n is the number of centroids between any two rows, m refers to the Received Signal Strength Indicators (RSSIs) acquired at the offline stage, and 1 is the number of rows that holds some of the grid points), (2) NNV4 outperforms the NN, KNN and Path-loss based Fingerprint Localization algorithms (PFL) in terms of accuracy by approximately 29%, 13%, 22%; respectively, (3) NNV4 outperforms the NN, KNN and PFL in terms of precision by approximately 53%, 28%, 52%; respectively, and (4) NNV4 has a lower probability of positional error compared to those for the existing indoor localization algorithms.