Dynamic Fingerprint Benchmarking Model for WiFi Indoor Localization

Abstract

WiFi indoor localization is regarded as the most promising reliable technology for indoor localization. Its concept is based on building fingerprint and training a ma-chine learning model on it, then using it for predicting the location based on the measured received signal strength RSS. One aspect of its vulnerability is its time vari-ant change due to the dynamical conditions in the environment and the access points APs. This was the most focus of the researchers in the last decades. However, minori-ty of them have considered the limitations of the existing benchmarking fingerprint in terms of lacking of support to realistic navigations runs when they are used for eval-uating indoor localization models. This article proposes a simple yet novel add-on to existing benchmarking fingerprint that assists in converting it from static fingerprint to consecutive training data with supportability to incremental learning and awareness of various characteristics of indoor navigation runs. The model was evaluated based on two benchmarking fingerprint and the generated incremental data were plotted with two levels of granularity one is corresponding to location and the other to place for referring to high resolution level and low resolution level.