Localização Indoor por Meio de Aprendizagem de Máquina Apoiada por Beacons Virtuais

Abstract

This article presents a solution to the indoor location problem through machine learning with the support of a new concept called virtual beacon. This concept has shown considerable performance gains in models where the representativeness of the data is crucial in the accuracy of the model's predictions. Virtual beacons can also be useful in environments where the installation of reference beacons at certain points could cause disturbance to the movement of people and objects in general. By way of performance comparison, the solution was implemented considering four different machine learning algorithms, two of them linear and the other two non-linear. Validations with real data pointed the model based on Multilayer Perceptron (MLP) as the model with the best performance among the four models considered with regard to the smallest error between the predicted and the real position, and the application of the virtual beacon concept outside determinant for such a result.