Fingerprint Augmentation Based on Location Information Consistency in Dynamic Environment

Abstract

The complex and dynamic structure of the indoor environment poses challenges for high-precision fingerprint localization. Changes in the indoor environment shift the data distribution domain of fingerprint data, leading to differences in data distribution between offline fingerprint databases and online data and reducing positioning accuracy. To address the time-consuming and labor-intensive task of reacquiring the database, we propose a method based on location information consistency for fingerprint data augmentation in dynamic environments. We treat fingerprint data of different changed environments as multiple domains and separate them into the location feature and the spatial feature by an encoder. The location feature represents the same statistical feature in different fingerprint data distributions, which is domain-invariant. The spatial feature corresponds to the unique statistical characteristics and random variations of each data distribution. Then, the generator is utilized to recombine the location feature with randomly sampled spatial features from the target spatial condition domain, generating diverse fingerprint samples of the target spatial conditions. Experimental results demonstrate that our method can effectively expand fingerprint data in dynamic environments and generate diverse outputs.