A Low Cost ECG Biometry System Based on an Ensemble of Support Vector Machine Classifiers

The electrocardiogram (ECG) found many clinical applications. Recently, it was proposed as a promising technology also for biometric applications, i.e., to recognize a subject within a group of known people. For such an application, the accuracy of classical ECG clinical recordings is usually not needed, but the measurement procedure should be fast, robust and cheap. We developed an embedded wearable system for recording one-lead ECG from the wrists of a person. The system was used to record data from 10 subjects. Data were pre-processed to reduce the noise content. Then fiducial points were detected and used to train an ensemble of support vector machines to identify a person among the group. Mean classification accuracy was higher than 95 % if a single heartbeat was considered and higher than 98 % if 3 consecutive heartbeats were used, choosing by majority. The system is fast (a few seconds are needed), not invasive and can be used either standalone or together with other identification techniques to increase the safety level.