Towards an indoor navigation system using Bluetooth Low Energy Beacons

Indoor navigation, the ability to find a path to an office, a conference room, or the exit in an unfamiliar building, is one of the most well-known applications of indoor positioning technology. This technology remains relatively underdeveloped due to the inherent difficulty of geolocalization in an indoor environment. Currently, we are working in an indoor navigation system based on Bluetooth Low Energy (BLE) Beacons fingerprinting and we created a testbed deployment of 30 beacons in the Center for Information Technologies of ESPOL University. This paper presents a comparison study of three machine learning techniques used to improve BLE fingerprinting accuracy within our testbed. Preliminary results show that Random Forest, 30% more accurate than Naive Bayes, is able to correctly estimate the location of multiple users with room-level accuracy 91% of the time.