Ambient intelligence pursues the integration of intelligent approaches on
an IoT infrastructure, mainly using everyday objects of the environment.
The main hypothesis of the work is that the way in which a user interacts
with a door handle is suitable to be used in the identification task. Our proposal contributes with a new method to identify persons in a seamless and
un-obstrusive way, suitable to be used in a smart building scenery without
the need to bring any additional device. In this case, we embed accelerometers
and gyroscopes in a door handle in order to obtain a data set comprising
samples of 47 individuals. A parametric approximation is adopted to reduce
each sample to a feature vector by using a dynamic time warping technique.
A study has been made of the outcomes of different classification techniques
over six different feature sets in order to assess the feasibility of this identification challenge. The AUC values observed with the selected feature set
show promising results above 0.90 using neural networks and SVM classifiers.
