Graph-Based WiFi Fingerprint Augmentation in Indoor Localization

Fingerprint-based WiFi positioning is a mainstream indoor localization method that requires a dense fingerprint database through extensive data collection at numerous reference points (RP). To mitigate this costly and labor-intensive process, this work proposes a graph-based dual channel fingerprint augmentation model. Specifically, a spatial graph is constructed where nodes represent WiFi fingerprints collected at RPs and edges are weighted by spatial distances to capture physical signal correlation. The model then encodes the WiFi fingerprint features and the RP adjacency structure with two distinct channels. A variational autoencoder (VAE) is employed to learn latent feature representations, while a generative adversarial network (GAN)-based discriminator is introduced to enforce consistency of the distribution between the real and generated WiFi fingerprint. Experimental results demonstrate that our model significantly reduces manual effort while achieving high localization accuracy even at a 40% data compression rate. And the proposed augmentation method outperforms the baselines by 11.7% in localization accuracy.