| Abstract: |
The spaces we live in are becoming increasingly smarter due to the integration of more and more sensors that acquire a large amount of information not only on the environment, but also on the people living in it. In this context, human identification is important not only to know who the collected data corresponds to, but also to guarantee that only authorized people can access the data and that information presentation is adapted to the user. In the scope of ongoing research projects aiming at health and well-being monitoring in smart homes, the main objective of this contribution is to explore the feasibility of user identification though non-invasive sensors, namely photoplethysmography (PPG) sensors. We propose a solution based on transfer learning and features corresponding to images (scalograms) extracted from the PPG signals to obtain a model for user identification. The results of the model’s evaluation, using a dataset obtained from ten subjects (three acquisitions each), show that an acquisition-independent solution can be challenging, even when considering a small interval between acquisitions, achieving a mean accuracy of 73% and mean F1 score of 72% (two acquisitions for training and a single window for classification). Nevertheless, these results show that there is potential in using a low-cost PPG sensor for biometrics in smart environments, especially if identity is decided based on several consecutive windows. This solution has different possible applications related to human monitoring, access control (e.g., rooms, devices, appliances), and adaptive interaction (e.g., adaptation of the output to the current user). |