An RSS Pathloss Considered Distance Metric Learning for Fingerprinting Indoor Localization

The immense diffusion of smart mobile devices increase the usage of WiFi as a state-of-the-art wireless communication standard dramatically, and it also renders fingerprinting method for high-accuracy indoor localization possible. However, the fundamental challenge of existing fingerprinting localization is that the complicated indoor spatial structure brings great difficulty to exploit the relationship between received signal strength (RSS) distribution and the process of fingerprint matching. To address this problem, in this paper, we focus on integrating the signal pathloss model into distance metric learning, and a novel cost similar to the Large Margin Nearest Neighbor (LMNN) is designed to conduct offline metric learning procedure. The proposed metric learning, which is pathloss model practically considered to learn the mapping between signal strength distribution and physical space structure, aims to obtain a reasonable distance matrix for k-Nearest Neighbor (KNN) based indoor localization. Experiment results corroborate that the proposed metric learning method can further improve the indoor localization accuracy compared with the conventional methods and its effectiveness is validated extensively in various cases for testing.