“…A set of them, in particular the first studies, proposed systems based on the extraction of a set of fiducial temporal and amplitude features from the ECG, from the P-QRS-T [1, 5-7, 10, 11, 16] or only the QRS-T segment [8,23], which is generally a difficult task. To bypass it, more recent approaches compute non fiducial parameters between windowed ECG into single heartbeat signals, needing only the R-peaks detection [13,19,21,22], except in [9,12] where no waveform detection is required. When the ECG is segmented, the template matching is performed between windows of various lengths: 100 ms in [19], 120 ms and 600 ms in [13], 700 ms in [21] and 750 ms in [22].…”