Path distance-based map matching for Wi-Fi fingerprinting positioning

Abstract

Map matching is a commonly-used technique that employs spatial constraints to improve positioning results. While map matching can improve the positioning performance to a large extent, existing map matching methods consider only adjacent transitions between reference points (RPs). This makes these map matching methods depend highly on the sampling size of RPs. To reduce the influence of the RPs’ sampling size, a novel map matching method called PDMatching is proposed in this paper, which considers both adjacent and non-adjacent transitions. These transitions are described based on the path distance of the RP sequence obtained by the shortest path algorithm. Compared to the commonly-used Euclidean distance, the path distance is more suitable for map matching as it takes into account spatial constraints. It allows to estimate the transition distance more accurately, which can further improve the positioning accuracy. To infer the location of a user, the student’s t-distribution is used to transform the path distance into a transition probability, from which the location can be obtained via the Viterbi algorithm. Extensive experiments have been conducted to evaluate the proposed PDMatching in a large museum environment. Experimental results show that the proposed PDMatching can achieve a mean localization error of 3.4m and 4.6m for uniform and varying speed modes, respectively, which outperforms the state-of-the-art methods (e.g., MapCraft, VTrack, XINS). Moreover, the PDMatching is more robust to the sampling size of RPs than other methods.