Gaussian Process Assisted Fingerprinting Localization

This paper presents an application of the firefly algorithm (FA) to Gaussian process (GP)-based localization. Partial radio-frequency (RF) signature map is first collected and used to train the GP model. The hyperparameters of the GP prior model are searched by the FA. GP regression is then used to generate an estimation of the RF signature map for the entire area to be localized. This is in contrast to traditional fingerprinting-based localization where a database of RF signature has to be collected for the entire area of interest. Using the estimated signature map, the position of the device is estimated using a combined likelihood function from multiple access points (APs). The proposed scheme relies on only existing infrastructures and can be used both indoor and outdoor. Experiments using indoor WiFi APs show that median error is around 3 m.