Hierarchical KNN for Smartphone-Based 3D Indoor Positioning

Abstract

Fingerprint-based localization, or positioning technique, is well-known to achieve high accuracy in location estimation in indoor environments where the multipath fading effect is severe. However, the accuracy of location estimation depends on the choice of the pattern matching techniques that are developed in the on-line phase. This paper proposes a new algorithm called Hierarchical K-Nearest Neighbors (KNN) for the pattern matching phase to estimate the location of the target in 3-dimensional (3D) indoor environments. For practical usage and saving budget and time for implementation, the Wi-Fi-based indoor positioning system (IPS) is implemented, and the smartphone is used as the user device. In this work, an Android smartphone is used for the study case. The results demonstrate that Hierarchical KNN achieves the lowest mean distance error (MDE) of approximately 3.263 m, outperforming various fundamental machine learning approaches such as Random Forest and KNN classifiers, with MDE reductions of 8.19% and 11.52%, respectively.