Um sistema de localização indoor usando fingerprinting e detecção de novidades para avaliação de confiança

Abstract

Indoor localization systems are used to locate mobile devices inside buildings where traditional solutions such as Global Navigation Satellite Systems (GNSS) do not work well due to the lack of direct visibility from satellites. Fingerprinting is one of the most known and accurate solutions for indoor localization, it is divided into two distinct phases: a training phase (Offline) and a localization phase (Online). One of the most used information is the Received Signal Strength Indicator (RSSI), as it is easy to obtain, but RSSI values are known to be unstable and noisy due to obstacles and the dynamics of the scenarios, causing inaccuracies in the estimates of position. Due to this RSSI characteristic, several methodologies have been proposed to try to mitigate the noise effects in the training phase, however, the RSSI information also presents noise in the localization phase. This noise often causes the system to point to a location that it is not sure is correct, even though it is most likely based on its calculations. To minimize this problem, this work presents some methods to verify the confidence level of classifications using classification probabilities combined with novelty detection algorithms. Thus, in this work, we propose LocFiND (Localization using Fingerprinting and Novelty Detection), a solution based on fingerprinting that uses novelty detection to assess the confidence of the estimated positions and, thus, try to mitigate the noise caused by RSSI on localization. Unreliable estimations are discarded and not forwarded to the application. We implemented our solution in a real-world, large-scale school area using Bluetooth-based devices. Our performance evaluation shows considerable improvement in the localization accuracy and stability while discarding only a few, unreliable estimations.