Documente Academic
Documente Profesional
Documente Cultură
Kuncoro Triandono Mukti1, Desti Mina rahayu2 and Anggunmeka Luhur Prasasti3
Computer Engineering, Faculty of Electrical Engineering,
Telkom University
Bandung, Indonesia
1
kuncoroteem@gmail.com, 2 destyviola@gmail.com, 3 anggunmeka@gmail.com
Photoplethysmography is a noninvasive
optical technique [7] to measure changes in blood
volume based on variations in the intensity of light
passing or reflected by human organs. Light
emitted in body tissues can be absorbed by
different substances, such as skin pigment, bone,
arterial blood and veins. Changes in blood flow are
most prevalent in arteries and arterioles [6]. It's just
that PPG Technique is quite sensitive to motion
artifacts, in research [13] proposed Techniques to
reduce the effects of motion artifacts using
accelerometers. Several other studies use digital
cameras and PPG Techniques to detect human
resources by using normal ambient light as a source
of illumination [5,9,14,15,16]. PPG is also used on
smartphones to monitor human physiology [17,18]. Figure 3. Variation of Light Absorption by Body
Network [6]
Photoplethysmography uses light source from
LED (Light Emitting Diode) and PD (Photo
considerable distance (> 1m) using ambient light
[6].
k=0
( n=0: N −1 )
Figure 7. Photodiode
�[�] = Matrix of PPG data in frequency
D. Independent Component Analysis domain, � = DFT size (DFT data size),
(ICA), a statistical technique used to �[�] = PPG data matrix in time domain, �
reveal independent source signals from a = Index of PPG data in frequency domain,
set of observed mixtures, used to complete � = Index of PPG data in time domain.
the separation of blind sources (BSS) [20, II. RELEATED WORK
21]. BSS aims to recover an unobserved
signal from a set of observations assumed In the study [4], a study of the correlation
as a linear mixture of some underlying between HR values obtained from two different
source. The "blind" property of the BBS smartphones (Droid and iPhone 4s) with HR values
relies on the fact that it is unknown that obtained from ECG signals. The analysis was
they are mixed. "Cocktail party problem " performed using Pearson Correlation method. The
can interpret BBS very well [22]. In this results show that there is a linear relationship
model, a group of microphones revolved between the measurements of the PPG-based
around the cocktail party room, where signal-based HR with the measurement of HR-
people talked together, noting the sounds based ECG signals. The results of the study stated
that were a mixture of people in the room that the smartphone can be used as a real-time HR
(see figure 8). BBS is used to separate measurement tool.
sound from all other microphones, to
eliminate pollution from other people's In the study [5], the authors discussed how
voices, and altogether according to the facial video images taken using a webcam could be
original sound. used in heart rate measurements. In general, the
system in this study will extract the
Photoplethysmography information from the facial
video image input through analysis and
postprocessing using the emguCV library to detect
faces. determines the Region of Interest (ROI)
measurement, and the ROI separation in each frame
into 3 color channels ie R (red), G (green), B
(blue). Then each channel is processed using the
Independent Component Analysis method to obtain
the heart rate as a result of measurement. The result
will be compared to a digital pulse oximeter, a
standard physiological signal gauge based on
photoplethysmography to determine the accuracy
of the measurement results. Post-processing and
Figure 8. The cocktail party problem analysis of video and physiological recordings is
done using software created using C #
E. Fast Fourier Transform (TTF), The programming language and development tools
Fourier transform (FT) is a mathematical Visual Studio 2010 Ultimate (Microosoft Corp.).
transformation used to change signals
then used an automatic face tracker to detect faces
between time domain and frequency
domain [12]. In practice, transformations in the video frame and localize the measurement
often occur in separate signal samples, and area (Region of Interest = ROI) for each video
frame. In the implementation phase, the emguCV
library is used to obtain the coordinates of the face
location [19]. To prevent facial segmentation errors
from affecting the performance of the algorithm,
the face coordinates of the previous frame are used
if no face is detected. Then the ROI is separated
into three channels and calculated the average of
each channel to get red (R), blue (B) and green (G)
measurements for each frame. Then Fast Fourier
Transform (FFT) is implemented on the source
signal to get the power spectrum. To overcome the
noise occurring in the pulse frequency calculation Figure 8. Example of research data retrieval results
the historical estimation process of the pulse [9]
frequency is used to reject the artifact by setting the
threshold for the maximum change in pulse 2. Data extraction, Extraction of data is done in
between consecutive measurements (taken 1 the form extract the feature on the video frame
second separately). Based on the results of the and convert it into a signal. The features used
study showed that the results of heart rate are color changes that indicate changes in
measurements produced with ICA approach the blood volume on the microvascular as a result
measurement results using pulse oximeter as a of human heart rate activity. The data
reference. extraction process in this study contains three
stages: determination of Region of Interest
In this study [9], this research will be (ROI), frame extraction, and conversion from
conducted related to detection of pulse using pixel value to signal wave. Region of Interest
camera smartphone which aim is to easy (ROI) is used to locate the image portion
monitoring health condition and can be used by containing the most significant color change
every person. The techniques and calculations used feature [5]. ROI in this research is done by
are photoplethysmography and discrete fourier scanning on one part of the specified frame.
transform (DFT). The image in the selected frame is divided into
4x3 blocks. Each block is calculated on the
average value of pixels (P ̅). The block that has
the most heterogeneous pixel value is indicated
by the value of P ̅ closest to 255/2. Blocks that
meet these criteria are selected as the video
extraction reference area. Then do a scan on
each block to find the part that contains the
most heterogeneous pixel value. It is used as a
reference area for extraction as shown in
Figure 7. Block diagram of the study [9] figure.9.
There are several steps done in this study as can be
seen in Figure 7, namely:
Noise