Uma abordagem para localização em ambientes internos baseada em impressão digital do sinal de Wi-Fi e aprendizagem de máquina

Abstract

Indoor Location Based Services (ILBS) have attracted a lot of attention in recent years because of their social and commercial potential, with an estimated market value of $10 billion by 2020. As satellite (GPS) and cellular (GSM) can’t penetrate indoors, with external walls and internal obstacles, the use of the Wi-Fi signal, for internal location, has gained a lot of importance in both academic and industrial areas due to the large penetration of wireless LANs (WLANs) and Wi-Fi-enabled mobile devices. In particular, the fingerprint of the received signal strength (RSS fingerprint) has attracted a great deal of attention by alleviating the multipath problem (multiple paths traveled by the signal between the sender and receiver) aggravated by the existence of walls and indoor objects. Given this scenario, motivated by the search for an efficient and effective ILBS, we propose the combined use of Wi-Fi fingerprint with Machine Learning algorithms in order to identify the approach more suitable for indoor localization and rational use of resources in mobile devices, considering the average error estimate of about 1 (one) meter. Unlike other related works, which concentrate efforts only on one of the phases of the RSS fingerprint, our research sought improvements both in the construction phase of the environment signal map (offline phase) and in the phase of locating the mobile device in the indoor environment (online phase). In this sense, one Design of Experiments was carried out with two scenarios (2D and 3D) and the results obtained showed a minimum average error of 2m in the 2D experiment, with only 5 measurements in the offline phase, and 1.08m in the 3D experiment, using a Neural Network and information from the previous position in the online location phase.