With the growth of the area in digital health, wearable devices stood out due to their practicality and comfort in detecting the personal data of their users. In general, these devices have a variety of sensors that capture information about the environment and user, heartbeat, amount of steps, oxygenation in the bloodstream, and photoplethymram, and electrocardiogram. Typically, capturing signals through sensors presents problems that impair signal analysis such as noise, false electromagnetic waves, and unexpected user movements. From this, signal filtering becomes an indispensable step in the process of removing these noises. The sensors can capture these signals, and through later steps, filtering, classification, one can reach in the identification of users. There are traditional methods of recognizing people, such as iris, face, or fingerprints. Today, biosignals are already able to be used for authentication, going through several steps to achieve this goal. Steps such as capture, filtering, extraction of characteristics, classification, and correlation are the main equipment used by these biosignals but still depend on mobile devices. There are many biosignals to be used in the authentication of people. Still, not all are effective; the choice is due to the fact of the shape of the graph of the generated signal, the quality in the capture of signals, and their ability to extract unique characteristics and classification. With these aspects pointed out, this master’s thesis aims to present a Double Identification System of wearable device users through the signs of Photoplethysogram and Electrocardiogram. With this, a method of parameterizing filtering and extracting peaks and the second method of double authentication through the biosignals PPG and ECG. The results indicated a correlation of 80% between the entire raw signal and filtered signal until peak extraction. They obtained an accuracy of 94.1% for the PPG signal without the calculation of errors and 99.98% with the calculation of the error rate, ECG signal reached 88.79% accuracy, for a total of 2809 inferences for each signal for identification of wearable device users.
|
