Data-Efficient Gaussian Process Regression for Accurate Visible Light Positioning

In the field of indoor localization systems, Received Signal Strength (RSS) based Visible Light Positioning (VLP) has gained increased attention due to the dual functionality of lighting and localization. Previously geometrical models have been used to determine the position of a mobile entity, however these are unsuited when dealing with tilted surfaces and non-Lambertian sources. For this reason, machine learning techniques like Multi Layer Perceptrons (MLPs) have been considered recently. In this work, Gaussian Processes (GPs) are introduced in the context of RSS-based VLP, since they have proven to work well when using small, noisy datasets for different applications. Their performance is evaluated using both simulated data with a small transmitter tilt tolerance and measurements. It is demonstrated that the GP model outperforms both the multilateration approach and the MLP approach for the simulations and measurements data.