A Comparative Study of Deep-Learning-Based Semi-Supervised Device-Free Indoor Localization

Real-time device-free indoor localization is a key technology for many Internet of Things (IoT) applications. Fingerprinting-based localization schemes rely on the database constructed from an offline site survey. Constructing a fully labeled database is expensive, and therefore a fingerprinting-based method requiring only a small amount of labeled data and a large amount of unlabeled data (i.e., semi-supervised) is highly useful. In this paper, we consider semi-supervised fingerprinting techniques based on the classic, generative model-based variational auto-encoder (VAE) and generative adversarial network (GAN). We conduct a comparative study of VAE and GAN in three real-world environments. Experimental results reveal that GAN generally outperforms VAE with various amounts of labeled data. Insights into how different generative mechanisms of these schemes, as well as environmental effects, affect the performance are provided.