TELKOMNIKA
Low-Cost Based E C++ and OpenCV library. The
the following sections.
2.3.1. Detection of the cent
The pupil may be dar contrast is sufficiently large. Y
algorithm to locate the pupils constraints using a skin-color
model and it will fail in the pre Yang et al applied the ellipse
coordinates of pupil edge pixe This research use simple tec
labeling.
The steps to detect th begins by capturing a eye ima
grayscaling process convert th smoothed using Gaussian filt
reduce sharp edges, aiding th
a
d Figure 6. Detection of the cen
c Gaussian blur image, d The next step is thresholding
clearly visible as a black c component as well. These rep
color as the pupil. The conn components in the image an
component is defined as a gr component labeling algorithm
Once all connected parameters of each blob such
are compared to experimenta discarded based on them. Du
will never be more than one Thus, the system selects the
center coordinate of this conne ISSN: 1693-6930
Eye Tracking and Eye Gaze Estimation I Ketut Ge he intricacies of these subsystems will be described
enter coordinate of the pupil
arker than their surroundings and thresholds may . Yang et al and Stiefilhagen at al introduce an i
ils by looking for two dark regions that satisfy certa lor model. Their method is limited by the results
resence of other dark regions such as eyebrows a se fitting algorithm to fit a standard ellipse or cir
ixels. The center of the ellipse or circle is the cente technique to detect the pupil center using conne
the center of pupil is used here are shown in Figur mage frame from the camera via OpenCV’s camera
t the RGB color to gray color space. After that the g filter to remove any noise in image. The smooth
the pupil detection system.
b c
e f center of pupil, a capture frame from camera, b g
d binary image, e component labelling image, f g the smoothed image to obtain the binary image.
component in this image. However, there may represent areas of the image of almost exactly the
nnected component labeling step exists here to and determine which is representative of the pu
group of pixels with values within a certain range m used is an open source add-on for OpenCV.
d component have been located, the system c uch as area, aspect ratio, roundness, and more. T
tally determined values for a pupil and connected Due to the nature of the human eye and surroundin
e connected component that fits all parameters f e correct connected component. The algorithm th
nnected component.
Gede Darma Putra 381
ed in more detail in
ay be applied if the n iterative threshold
rtain anthropometric lts of the skin-color
s and shadows [10]. circle based on the
nter of the pupil [7]. nected components
gure 6. The process era subsystem. The
grayscale image is thing also helps to
grayscale image, , f output image
e. The pupil will be ay be other black
he same shade and to locate all black
pupil. A connected ge. The connected
calculates several . These parameters
d componentes are ding features, there
s for a human eye. then calculates the
TELKOMNIKA
Vol. 9, No. 2, 382
2.3.2. Calibration and Poin