Improving Indoor Localization Through Data Augmentation of Visualized Multidimensional Fingerprints via Enhanced Generative Networks

Abstract

Fingerprint-based localization utilizes wireless signals to sense indoor environments, which has attracted significant research attention due to its advantages of wide deployment and low cost. However, attaining precise localization demands tremendous high-density and high-availability fingerprint measurements, which makes the site survey for signal collection pretty time-consuming and labor-intensive. To address this challenge, this article proposes a framework called augmented visualized fingerprint-based localization (AVF-Loc), which visualizes multidimensional wireless signals as fingerprint images and implements data augmentation to improve positioning in indoor environments. It begins by converting multidimensional wireless signals into low-resolution (LR) fingerprint images. Then, it employs an enhanced super-resolution generative adversarial network (ESRGAN) to realize data augmentation, which is designed to reconstruct the LR images into the corresponding high-resolution (HR) images. Subsequently, these HR images are transformed back into an augmented fingerprint. Based on the augmented data, the k-means weighted k-nearest neighbor (WKNN) algorithm is implemented for localization. Real and simulated experiments with 5G synchronization signal block (SSB) and Wi-Fi were conducted to evaluate AVF-Loc’s performance. The results indicate that AVF-Loc has significantly enriched the number of fingerprints and improved localization accuracy in real tests by 19.1056% and 13.3254% for 5G SSB and Wi-Fi, respectively, while by about 33.3082% in simulated experiments. Moreover, it outperforms state-of-the-art methods well. Extended analysis displays that AVF-Loc performs outstanding scalability and robustness in complicated indoor environments. Overall, the proposed AVF-Loc is demonstrated to have superiority in improving indoor localization through ESRGAN-based data augmentation of visualized multidimensional fingerprints.