Methods
The individual color signals (Figure 2a) were
detrended and smoothed with a 5-point moving
-
cant sources of noise (Figure 2b). A 0.5 Hz – 3.7
< $
>'
%
(
(
(
raw signal (Figure 2c), and independent compo-
(
using a FastICA program in Matlab (Gavert et al.,
'??@" 69 %
-
( 5%
iteration to solve for the source signals (Hyvärinen
J QX '???" Y
(
each signal were not included in the ICA. In each
case, one of the resulting source signals was found
to consist of regular oscillations resembling heart
beats (Figure 3).
Fast Fourier transforms (Figure 4) were performed on
each of the source signals and the amplitude of the
highest peak within 0.5 and 3 Hz was recorded for
each transform. The source signal with the greatest
! (
-
senting cardiac rhythm. The frequency at which this
peak occurred was taken as heart rate, in beats
per second. Heart rate was also calculated from
the portion of the ECG recording corresponding to
the portion of video data analyzed, using a simple
Matlab program which detects peaks in the ECG
and the time interval between them. Video acquisition
and raw RGB signal generation were completed
in Labview; signal processing, ICA, and FFT analysis
were performed using Matlab