Gaussian Processes for Received Signal Strength Based Device-free Localization

In this paper, data-driven Gaussian process models for received signal strength-based device-free localization are presented. The models have physically interpretable parameters that can be explained by analytical models. Moreover, the presented models can approximate highly nonlinear signal propagation patterns such as a target's influence on existing multipath components which is generally treated as unwanted noise that degrades the system performance. The models are evaluated with experimental data and the results indicate that compared to a prior-art reference model, the proposed models decrease the modeling error and improve the localization accuracy, especially in multipath rich indoor environments.