=Paper=
{{Paper
|id=Vol-3013/20210158
|storemode=property
|title=Automated Monitoring of Content Demand in Distance Learning
|pdfUrl=https://ceur-ws.org/Vol-3013/20210158.pdf
|volume=Vol-3013
|authors=Viktor Shynkarenko,Valentyn Raznosilin,Yuliia Snihur
|dblpUrl=https://dblp.org/rec/conf/icteri/ShynkarenkoRS21
}}
==Automated Monitoring of Content Demand in Distance Learning==
https://ceur-ws.org/Vol-3013/20210158.pdf
Automated Monitoring of Content Demand in
Distance Learning
Viktor I. Shynkarenko, Valentyn V. Raznosilin and Yuliia Snihur
Dnipro National University of Railway Transport named after Academician V. Lazaryan, 2 Lazaryana str., Dnipro,
49010, Ukraine
Abstract
In this paper the research of means and the development of software for matching the student’s gaze
focus with the structure of information on the computer monitor during distance learning is presented.
Widespread hardware is envisaged to be used. Primary processing of the face image, eye regions
separation is performed by means of the OpenCV library. An appropriate algorithm to calculate the
center of the eye’s pupil has been developed. The influence of the system calibration process with
different schemes of calibration point display, its delay time on the screen and location of the additional
camera according the accuracy of the calculation the coordinates of the gaze focus is investigated. Based
on the performed experiments, it was defined that the error of gaze focus recognition with using two
cameras can be reduced to 4-10%. The proposed approach makes it possible for objective measurement the
working time of each student with one or another part of content. The lecturer will have the opportunity
to improve the content by highlighting significant parts that receive little attention and simplifying
those elements that students process for an unreasonable amount of time. It is planned to integrate the
developed software with the LMS Moodle in the future.
Keywords
Distance learning, educational content, program tools, oculography, gaze focus
1. Introduction
Distance learning is becoming more popular as higher education develops and modernizes. This
is made possible by improved technical capabilities (computer public and private networks,
digital knowledge bases, and so on) and the emergence of specialties that can be mastered
remotely, without direct contact between the student and the lecturer. IT, finance, management,
and other specialties are among them.
Aside from the obvious benefits of distance learning methods, there are some significant
drawbacks:
• the learning process monitoring is absent or seriously difficult (attention, interest, under-
standing of the material by the student);
ICTERI’21: 17th International Conference on ICT in Education, Research and Industrial Applications, September 28 –
October 02, 2021, Kherson, Ukraine
" Shynkarenko.vi@gmail.com (V. I. Shynkarenko); valentin.raznosilin@gmail.com (V. V. Raznosilin);
snigurjulia150498@gmail.com (Y. Snihur)
0000-0001-8738-7225 (V. I. Shynkarenko); 0000-0002-4463-4588 (V. V. Raznosilin); 0000-0001-7294-6821 (Y. Snihur)
© 2021 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073 CEUR Workshop Proceedings (CEUR-WS.org)
• the possibility of an objective knowledge assessment has been complicated (person
authentication, self-fulfillment of tasks, use of only permitted materials).
The first drawback is that modern students have difficulty concentrating when they master
the material on their own. Clip thinking, caused by an abundance of unstructured information
and its ease of access, has a negative impact.
The second relates to the evolution of communication tools, high-speed information retrieval,
which can be used in the testing and examination process in the absence of a lecturer.
A method and corresponding software toolkit for tracking the gaze focus on the training
material elements of the training course are presented in this article. A necessary requirement
for development is the solution of the problem using the bare minimum of technical means to
ensure mass application.
As a result, it is possible to objectively measure what the student pays more attention to,
what is ignored, how much time is spent studying this or that section of the teaching material,
and so on.
Based on the data obtained, it is planned to develop tools for analyzing the process of students’
work with teaching materials using big data methods. In this case, both individual and group
work patterns of various user groups can be considered (based on filtering by age, gender,
level of initial training and other criteria). Finally, this will allow for a reasonable correction of
training materials in order to improve the efficiency of their perception.
Assumptions are made as follows:
• a student looks at the content on a laptop screen with a resolution of 1366x768 pixels and
physical dimensions of 340x200 mm;
• the built-in laptop video camera is located at the top of the screen and in the center
horizontally, with the option of using an additional video camera;
• the distance from the screen to the bridge of the nose is 500 mm;
• the student sits almost motionless, centered on the screen;
• the average distance between the pupils is taken equal to 64 mm;
• there are no significant (more than 10-15 degrees) turns and tilts of the head.
Under the conditions described above, the user’s face contour occupies about 50% of the
frame’s height and about 25-30% of the frame’s width. It’s allowing for fairly accurate recognition
of the face contour as a whole and its individual parts (eye contours, pupils, temples, bridge of
the nose, etc.).
2. Related works
To determine the focus point of the user’s gaze on the monitor screen oculography or eye-
tracking [1, 2] was used – a technology that allows you to observe and record eye movements:
pupil dilation, its movement, etc.
Eye tracking is an effective research tool in various fields of education. Using this method, it
is possible to analyze how the individual characteristics of students in combination with various
properties of the teaching material can influence the learning process [3]. Eye focus tracking,
in particular, enables to reveal [3]:
• student behavior features when selecting information and solving problems;
• differences in teaching strategies among different students;
• model of social interaction between lecturer and student;
• the effectiveness of training materials;
• what content elements attract and hold the student’s attention.
Industrial gaze capture systems exist [4, 5]. Contact lenses with built-in mirrors, infrared
illuminators that are reflected by the eyeballs and recorded by a video camera, and so on are
examples. However, all of these methods necessitate complex and costly hardware and software,
and they are inapplicable in the context of widespread use of distance learning.
The feasibility of developing a system for determining the gaze focus with a bare minimum
of hardware (for example, a standard video camera on a modern laptop) was investigated [6, 7].
It should be noted that the required measurement accuracy is not achieved in these works, and
there is no binding to the elements of the training content.
3. Research and project implementation
Computer vision is a set of software and hardware tools that read images in digital form and
process them in real time. Various specialized software libraries are used in the developing
computer vision systems [8, 9]. We chose OpenCV, which is a library of algorithms for computer
vision [10, 11], image processing, and numerous general purpose open source algorithms.
OpenCV is one of the most well-known libraries for dealing with computer vision issues. It
is written in C++, but it can also be used in Python, Java, Ruby, Matlab, and Lua, and it is
supported by a variety of platforms, including MS Windows, Linux, Mac OS, Android, and
iOS. The library includes more than 500 functions. In particular, there are many optimized
algorithms associated with the processing and analysis of video frames. Filtering, contour
search, geometric transformations, motion analysis, object detection, observation, and many
other functions have been implemented. It is possible to work with xml files.
The main stages of the gaze focus tracking process are highlighted:
1. Receiving video footage from a webcam that broadcasts a live image of a student’s face;
2. Pre-processing – preparation of the image (translation into black and white, increasing
contrast, etc.);
3. Facial feature recognition to detect the area of the eye on the video frame;
4. Eye recognition – tracking the contours of the eyes and the position of the pupils;
5. Calibration – comparison of the pupils position with the coordinates of a known point
on the screen, which corresponds to the gaze focus;
6. Calculation of the gaze focus coordinates on the screen based on the dependences obtained
in the previous stage.
The first three stages and partly the fourth are performed using the functions of the OpenCV
library. Stages one and two are engineering and are not of scientific interest. Let us go over the
third stage in greater depth.
Active appearance models (AAM), which are provided by the OpenCV computer vision
library, were chosen to recognize facial features because they are designed to accurately locate
anthropometric points on a facial image. AAM are statistical models of images that can be
adjusted to the real image by various transformations. The toolkit, which is based on an AAM,
is pre-trained on a set of pre-marked images.
There are two types of parameters used: shape-related (shape parameters) and statistical
image model or texture-related parameters (appearance parameters).
The shape is defined as a set of landmark points on the face [12]. They describe certain facial
features. Each landmark point defines a face morphological property and has its own number.
Figure 1 shows a similar markup.
Figure 1: Landmark points of the human face recognized by the OpenCV library
The image in the presented example shows 68 landmark points that form the shape of the
active appearance model. This shape represents the outer face contours, including the contours
of the mouth, eyes, nose, and brows. This markup allows to determine the various parameters
of the face in the image, which can be used for further processing.
For initializing the facial recognition method, the system uses a trained AAM that can be
used to predict facial features. This model is presented in the form of a faces database [13]. This
database includes variations of facial representation: different poses, lighting, etc.
3.1. Calculation of position of eye center and pupil
The fourth stage of the eye focus tracking system is the calculation of the coordinates of the
centers of the eye and pupil. The position of the eye center depends on the tilt and / or rotation
of the user’s head. The position of the pupil center depends on the rotation of the eyeball and
can change relative to the eye center, which will remain unchanged. At this stage, only the
region of the video frame containing the image of the eyes is considered. Each eye is processed
individually (fig. 2).
Let p1, .., p6 (points 36 –48 on Fig. 1) be the anthropometric points of the eye area. Each
Figure 2: An example of video frame parts that are used to recognize the contour of the pupil (a – left
eye; b – right eye)
point is given by a pair of x and y coordinates. Figure 3 and (1) demonstrate the calculation of
the coordinates of the eye’s center.
Figure 3: Eye scheme for calculating its center
The coordinates of the center of the eye (CoE) are calculated as following:
𝑝1𝑥 + 𝑝4𝑥 𝑝2𝑦 + 𝑝3𝑦 + 𝑝5𝑦 + 𝑝6𝑦
𝐶𝑜𝐸𝑥 = , 𝐶𝑜𝐸𝑦 = , (1)
2 4
where 𝑝1𝑥 , 𝑝4𝑥 , 𝑝2𝑦 , 𝑝3𝑦 , 𝑝5𝑦 , 𝑝6𝑦 – the coordinates of the corresponding points.
One of the most difficult steps in tracking the gaze focus is determining the position of the
eye pupil. The final result is significantly dependent on the values obtained at this stage, since
the coordinate of the gaze focus can be calculated by analyzing the displacement of the pupil
relative to the eye center. The process of tracking the eye pupil consists of several steps.
At the first step, the video frame is processed. The pupil is the part of the eye that is the
darkest. To recognize the pupil, need to increase the intensity of the colors and then convert
the frame to bi-tone.
To reduce the noise level median filtering is used. Depending on the aperture size (which
are square, central) of the median filter, the various sizes and the positions of the pupil can be
obtained (fig. 4).
Figure 4: The impact of aperture size on the final image
A threshold algorithm is used to highlight the darkest areas and change the image color
model from grayscale to bi-tone [14, 15].
The fig. 5 shows the dependence of the result from the threshold value. The successful
application of the threshold algorithm in this example is the use of T = 40, since the binary
image clearly shows the area of the eye pupil.
Figure 5: Dependence of the algorithm result from the selected threshold value (T)
The next step of eye pupil tracking is to find the contours of the iris on the binary image.
Contour – a curve that connects all points (along the border), that have the same color. The
Satoshi Suzuki algorithm [16], provided by the OpenCV computer vision library, is used to
recognize contours on a binary image. If there are several contours, the largest one is selected
because it corresponds to the iris of the eye. In fig. 6, examples of recognizing the elements
contours are shown.
Figure 6: Examples of iris contours recognition
The iris has the shape of a circle. To calculate the position of the eye pupil, need to find a
circle that fits and coincides with the contour of the iris obtained at the previous step. The
OpenCV library’s algorithm "The smallest enclosing disk" [17] is used for this purpose. Figure 7
shows examples of the iris circle and the eye pupil that corresponds to them.
Figure 7: Example of finding the eye pupil position
3.2. Calibration of the gaze focus coordinate on the monitor with the pupil
position
In the fifth stage, correspondence is established between the position of the pupil and a known
point on the monitor. This requires calibration. Calibration is a sequential demonstration on the
monitor screen of a point with known coordinates with simultaneous registration of the user’s
gaze directed at it. The movement of the calibration point on the monitor is depicted in fig. 8.
Figure 8: The movement of the calibration point
The program changes the position of the point after a certain period of time and calculates
the corresponding eye pupils coordinates. During calibrating, the student must keep a close eye
on the point on the monitor. To get rid of incorrect data obtained during blinking, the system
filters it out. The relationship between the width and height of the open eye is used to achieve
this. According to the study [18], the following is can be used to calculate the ratio of the sides
of the eye:
|𝑝1𝑦 − 𝑝6𝑦 | + |𝑝3𝑦 − 𝑝5𝑦 |
𝐸𝐴𝑅 = , (2)
2 |𝑝1𝑦 − 𝑝4𝑦 |
where 𝐸𝐴𝑅 – the ratio of the eye sides.
The ratio of the eye sides is almost constant as long as the eye is open and close to zero when
the person is blinking. Figure 9 depicts eye’s sides ratio graph. The EAR threshold is set at 0.15.
If the EAR is less than or equal to the threshold value, the system detects blinking and ignores
the corresponding frames.
Figure 9: Dependence of EAR on the state of the eye
The result of the calibration is an eye pupils coordinates array corresponds to the positions of
the calibration point. The next step is to process the obtained data – to find the linear regression
dependence of the pupil coordinates on the calibration point position on the screen:
𝐿 = 𝛼 · 𝐶 + 𝛽, (3)
where 𝐿 = 𝐶𝑜𝐸 − 𝐶𝑜𝐼, 𝐶𝑜𝐸, 𝐶𝑜𝐼 – coordinates of the eye and pupil centers, 𝐶 – is
the coordinate of the calibration point position, 𝛼 and 𝛽 – are unknown coefficients. The
linear dependence is calculated separately for the horizontal and vertical axes (𝑥 and 𝑦 coordi-
nates, respectively). The least squares method is used to calculate the parameters of the linear
regression:
𝑁 𝑁 ∑︀𝑁 ∑︀𝑁 ∑︀𝑁
1 ∑︁ 𝛽 ∑︁ 𝑁 𝑖=1 𝑙𝑖 𝑐𝑖 − 𝑖=1 𝑙𝑖 · 𝑖=1 𝑐𝑖
𝛼= 𝑐𝑖 − 𝑙𝑖 , 𝛽= ∑︀𝑁 2 ∑︀𝑁 , (4)
𝑁 𝑁 𝑁 𝑖=1 𝑙𝑖 − ( 𝑖=1 𝑙𝑖 )2
𝑖=1 𝑖=1
where 𝑁 – data sample size (𝑦𝑖 , 𝑥𝑖 ).
3.3. Eye focus calculation
Calculations are performed for each frame of the video stream except for frames where are
blinking detected. The coordinates of 𝐶𝑜𝐸, 𝐶𝑜𝐼 and their difference 𝐿 for the left and right
eye on both axes are determined (𝐿𝑙𝑒𝑓 𝑡,𝑋 , 𝐿𝑙𝑒𝑓 𝑡,𝑌 , 𝐿𝑟𝑖𝑔ℎ𝑡,𝑋 and 𝐿𝑟𝑖𝑔ℎ𝑡,𝑌 ). The average (for two
eyes) displacement value is calculated:
𝐿𝑙𝑒𝑓 𝑡,𝑋 + 𝐿𝑟𝑖𝑔ℎ𝑡,𝑋 𝐿𝑙𝑒𝑓 𝑡,𝑌 + 𝐿𝑟𝑖𝑔ℎ𝑡,𝑌
𝐿𝑎 𝑣𝑒𝑟𝑋 = , 𝐿𝐴 𝑣𝑒𝑟𝑌 = , (5)
2 2
where 𝐿𝐴 𝑣𝑒𝑟𝑋 – the average displacement along the 𝑥 axis, 𝐿𝐴 𝑣𝑒𝑟𝑌 – the average displace-
ment along the 𝑦 axis.
The coordinates of the gaze focus are calculated as next:
𝐿𝑎 𝑣𝑒𝑟𝑋 − 𝛼𝑋 𝐿𝑎 𝑣𝑒𝑟𝑌 − 𝛼𝑌
𝑃 𝑜𝐺𝑋 = , 𝑃 𝑜𝐺𝑌 = , (6)
𝛽𝑋 𝛽𝑌
where 𝛼𝑋 , 𝛽𝑋 – linear regression coefficients obtained at the calibration stage by the 𝑥 axis
and 𝛼𝑌 , 𝛽𝑌 – by the 𝑦 axis using (4). If the obtained gaze focus coordinates exceed the monitor
boundaries, they are given the nominal value of the monitor coordinate limits. When using
an additional camera, calibration is simultaneously performed separately for each camera and
dependencies (3) and (4) are determined. When processing a video image, the position of the
gaze focus (6) is calculated separately with subsequent averaging.
3.4. Matching eye focus with a content
A method for matching the user’s gaze focus with the information on the monitor is proposed.
Educational content (such as a tutorial chapter etc) is converted to a fixed-width bitmap. The
height of the image depends on the size of the content. The resulting image is divided into
regions that describe the structure of the content.
The student can view the content by scrolling the image up or down. For each video frame, the
program calculates the gaze focus point. The computation frequency depends on the parameters
of the video stream. The gaze focus is averaged over a specified time interval (for example,
within 1-2 seconds).
Figure 10 depicts an example of content splitting into structural elements.
Figure 10: An example of splitting content into separate elements
In figure 10 circles are used to highlight various types of elements:
1. title of the article;
2. drawing;
3. list;
4. a paragraph of the text;
5. software source code.
Next, a search is performed for the content element on which the gaze is focused. The
corresponding item identifier saves into the list. The resulting data can be used to visualize
how users interact with the content. For example, the program can reconstruct the sequence
of the user’s gaze movement; highlight content elements on which the gaze was focused for
the minimum or maximum time, and so on. In addition, it can calculate various statistical
parameters that characterize the users experience with this content.
The structure of the content is described in the form of a JSON file, which has the following
format:
content_structure ::=
{
"source": {
"title": "",
"type" : "< book | monograph | article | manual >",
"media": "< img | pdf | docx | web >",
"link" : "",
"image": "" },
"content": {
"heading" : ,
"text" : ,
"picture" : ,
"formula" : ,
"code" : ,
"list" : }
}
::=
[
{ "id":, "pos" : [,,,], "info":""},
...
{ "id":, "pos" : [,,,], "info":""}
]
::=
[
{ "id":, "pos" : [,,,], "info":"",
"items" : [,..,]},
...
{ "id":, "pos" : [,,,], "info":"",
"items" : [,..,]}
]
The description of an individual element consists of information about the frame enclosing
it: coordinates of the upper left corner, height and width. In addition, a free-form text label is
provided for each element. An example, for a drawing, the label can contain the number and
title of the drawing, for a paragraph, it can be empty or briefly describe its content, etc.
Additional information is provided to identify the position (vertical) of the list’s individual
items when describing an element in the form of a list (for example, references).
4. Results and discussion
Several experiments were performed to assess the impact of the number of points and the delay
in their demonstration during calibration, an additional camera, and slight head movements on
the accuracy of the results.
In the performed experiments:
1. uses a laptop with a screen resolution of 1366x768, a built-in camcorder with a resolution
of 1280x720. Experimental conditions: the timer of calibration point position change is
equal to 15 sec; changing the position of the calibration point is crosswise in the center
of the monitor, horizontally 9 points, vertically 5; the static position of the head;
2. unlike the previous one, an additional external video camera with a resolution of 1280x1024
is used, which is located at the bottom and in the middle of the monitor screen;
3. equipment is as second experiment. Experimental conditions: the timer of calibration
point position change is equal to 10 sec; changing the position of the calibration point has
the sawtoothed form (like a Fig. 6) horizontally 3 points, vertically 3; the static position
of the head;
4. equipment is as in the 2nd and the 3rd experiment. Experimental conditions: the timer of
calibration point position change is equal to 6 sec; changing the position of the calibration
point has sawtoothed form, horizontally 6 points, vertically 6; the static position of the
head;
5. equipment and conditions are as in the experiment 4, except for changing the position of
the head (rotations up to 10-15 degrees) when calibrating and calculating the gaze focus;
6. equipment and conditions are as in the experiment 4, except for changing the position
of the additional camera, which is located on the top and in the middle of the monitor,
screen.
Table 1 displays the results of the software’s accuracy under different conditions of use.
The results of the experiments demonstrated that using two cameras produce the best results
for tracking the gaze focus along the y-axis. It was also found that adjusting the calibration
process has a significant impact on the final result. The advantages of using two cameras are
clearly demonstrated in experiment 2. The error of the results for each camera is more than if
these two results are generalized (Table 2).
Table 1
Errors of gaze focus tracking in performed experiments
№ Error by the Error by the Error by the Error by the
horizontal axis (px) horizontal axis (%) vertical axis (px) vertical axis (%)
1 51 3,8 214 27,9
2 58 4,2 71 9,3
3 84 6,2 94 12,3
4 80 5,8 73 9
5 223 16 127 16
6 193 14 109 14
Table 2
Deviation of the calculated gaze focus coordinates from the coordinates of the calibration point
№ Error by the Error by the Error by the Error by the
horizontal axis (px) horizontal axis (%) vertical axis (px) vertical axis (%)
Camera (internal) 90 6,6 90 11,7
Camera (external) 67 4,9 87 11,4
For both cameras 58 4,2 71 9,3
As an example, the graphs of the calibration point coordinates on the screen (PoS) and the
calculated coordinates of the gaze focus (PoG) when tracking the user’s gaze using both cameras
(experiment 2) by the 𝑥-axis are presented in fig. 11 and fig. 12.
Figure 11: Coordinates of the calibration point and gaze focus
When the results of experiments where the user holds the head in a static or dynamic state
were compared, it was found that tracking of the gaze focus is better if the head is held in a
static state. It because the tracking system better recognizes pupil displacement under these
conditions.
Existing approaches are based on direct observation of the trainee or use costly equipment.
This presupposes conducting separate experiments to assess attention on structural elements of
Figure 12: The relative error of calculation the gaze focus coordinates
the studied material in each specific case.
The method presented in this work involves a massive study of the behavioral reactions of
students who are either in the classroom or at home and have standard devices like a laptop.
Unlike other works, the goal here was not to achieve pixel (or close to it) accuracy in determining
the gaze focus. For the purposes specified, an accuracy of 50-100 pixels along each axis of the
monitor is adequate. The specified accuracy corresponds to the size of the content (paragraph,
drawing, formula, etc.) with which the student interacts. Furthermore, an additional USB video
camera can be used, which on the one hand significantly improves accuracy (especially along
the vertical axis) while not reduce the target audience too much because such cameras are
widely available and relatively inexpensive.
5. Conclusions
The approach to the development of tools for tracking the gaze focus compliance with the
structure of information on a computer monitor is studied in this paper. Using hardware that is
available to everyone to improve the assimilation of information by students during distance
learning proposed.
Implemented software environment that is able to investigate the correspondence of the gaze
focus with the structure of information on the monitor using one or two cameras.
Several experiments were carried out under various conditions, including calibration process
settings, changing the position of the additional video camera, and the student’s head state
during the experiment. Based on the results of the experiments, it was concluded that the error
of gaze focus recognition is up to 10
The widely used in Ukrainian universities LMS Moodle distance learning system lacks
tools for assessing the significance, and intelligibility of teaching materials and their elements.
Comprehensibility is assessed indirectly through student testing.
The proposed method allows for the objective measurement of each student’s work time with
one or another element of content. The lecturer has the opportunity to highlight significant parts
that receive little attention and to simplify elements that students process for an unreasonable
amount of time.
This work was carried out in continuation of the research project of the tracking and analysis
of learning processes, in particular, to programming [19, 20, 21].
It is planned to integrate the developed software with the LMS Moodle in the future, as well
as to refine the algorithm for identifying the coordinates of the pupil and flare on the eyeball in
order to improve the accuracy of recognizing the gaze focus along the vertical axis in different
lighting and other conditions (for example, the presence of glasses, overlapping hairstyles).
References
[1] L. R. Young, D. Sheena, Survey of eye movement recording methods, Behavior research
methods & instrumentation 7 (1975) 397–429.
[2] F. Wadehn, T. Weber, D. J. Mack, T. Heldt, H.-A. Loeliger, Model-based separation, detection,
and classification of eye movements, IEEE Transactions on Biomedical Engineering 67
(2019) 588–600. doi:10.1109/TBME.2019.2918986.
[3] Understanding different aspects of learning, 2020. URL: https://www.tobiipro.com/
applications/scientific-research/education/, accessed 07 September 2020.
[4] S. Y. Gwon, C. W. Cho, H. C. Lee, W. O. Lee, K. R. Park, Robust eye and pupil detection
method for gaze tracking, International Journal of Advanced Robotic Systems 10 (2013) 98.
doi:10.5772/55520.
[5] C. W. Cho, J. W. Lee, K. Y. Shin, E. C. Lee, K. R. Park, H. Lee, J. Cha, Gaze detection by
wearable eye-tracking and nir led-based head-tracking device based on svr, Etri Journal 34
(2012) 542–552. doi:10.4218/etrij.12.0111.0193.
[6] E. Skodras, V. G. Kanas, N. Fakotakis, On visual gaze tracking based on a single low
cost camera, Signal Processing: Image Communication 36 (2015) 29–42. doi:10.1016/j.
image.2015.05.007.
[7] O. Ferhat, F. Vilariño, Low cost eye tracking: The current panorama, Computational
intelligence and neuroscience 2016 (2016). doi:10.1016/j.image.2015.05.007.
[8] Y. V. Bulatnikov, A. A. Goeva, Sravnenie bibliotek kompyuternogo zreniya dlya
primeneniya v prilozhenii, ispolzuyushchem tekhnologiyu raspoznavaniya ploskikh izo-
brazheniy (comparison of computer vision libraries for use in an application using flat
image recognition technology), Vestnik Moskovskogo gosudarstvennogo universiteta
pechati (2015) 85–91.
[9] G. Shakhin, Sravnitelnyy analiz bibliotek kompyuternogo zreniya (comparative analysis
of computer vision libraries), in: Colloquium-journal, 24 (48), 2019, pp. 53–55. doi:10.
24411/2520-6990-2019-10812.
[10] Y. Ji, S. Wang, Y. Lu, J. Wei, Y. Zhao, Eye and mouth state detection algorithm based on
contour feature extraction, Journal of Electronic Imaging 27 (2018) 051205 1–8. doi:10.
1117/1.JEI.27.5.051205.
[11] D. Chandrappa, G. Akshay, M. Ravishankar, Face detection using a boosted cascade of
features using opencv, in: International Conference on Information Processing, Springer,
2012, pp. 399–404. doi:10.1007/978-3-642-31686-9_46.
[12] Facial point annotations, 2020. URL: https://ibug.doc.ic.ac.uk/resources/
facial-point-annotations/, accessed 07 September 2020.
[13] Shape predictor 68 face landmarks, 2020. URL: https://github.com/davisking/dlib-models/
blob/master/shape_predictor_68_face_landmarks.dat.bz2, accessed 07 September 2020.
[14] Tracking your eyes with python, 2020. URL: https://medium.com/@stepanfilonov/
tracking-your-eyes-with-python-3952e66194a6, accessed 07 September 2020.
[15] Obrobka rastrovykh zobrazhen (raster image processing), 2020. URL: https://www.tobiipro.
com/applications/scientific-research/education/, accessed 07 September 2020.
[16] S. Suzuki, et al., Topological structural analysis of digitized binary images by border
following, Computer vision, graphics, and image processing 30 (1985) 32–46.
[17] E. Welzl, Smallest enclosing disks (balls and ellipsoids, in: New results and new trends in
computer science, Springer, 1991, pp. 359–370.
[18] J. Cech, T. Soukupova, Real-time eye blink detection using facial landmarks, Cent. Mach.
Perception, Dep. Cybern. Fac. Electr. Eng. Czech Tech. Univ. Prague (2016) 1–8.
[19] V. Shynkarenko, O. Zhevago, Visualization of program development process, in: 2019
IEEE 14th International Conference on Computer Sciences and Information Technologies
(CSIT), volume 2, IEEE, 2019, pp. 142–145. doi:10.1109/STC-CSIT.2019.8929774.
[20] V. Shynkarenko, O. Zhevago, Development of a toolkit for analyzing software debug-
ging processes using the constructive approach, Eastern-European Journal of Enterprise
Technologies 5/2 (2020) 29–38. doi:10.15587/1729-4061.2020.215090.
[21] V. Shynkarenko, O. Zhevaho, Constructive modeling of the software development process
for modern code review, in: 2020 IEEE 15th International Conference on Computer
Sciences and Information Technologies (CSIT), volume 1, IEEE, 2020, pp. 392–395. doi:10.
1109/CSIT49958.2020.9322002.