Evaluation of Thresholding Based Noncontact Respiration Rate Monitoring using Thermal Imaging.

International Refereed Journal of Engineering and Science (IRJES)
ISSN (Online) 2319-183X, (Print) 2319-1821
Volume 6, Issue 7 (July 2017), PP.50-58
www.irjes.com 50 | Page
Evaluation of Thresholding Based Noncontact Respiration Rate
Monitoring using Thermal Imaging.
*
Abdulkadir Hamidu Alkali
Department Of Computer Engineering.University Of Maiduguri, Borno State, Nigeria.
Corresponding author: *Abdulkadir Hamidu Alkali
Abstract: - A noncontact method for respiration rate monitoring using thermal imaging was developed and
evaluated. Algorithms to capture images, detect the location of the face, locate the corners of the eyes from the
detected face and thereafter locate the tip of the nose in each image were developed. The amount of emitted
infrared radiation was then determined from the detected tip of the nose. Signal processing techniques were then
utilised to obtain the respiration rate in real-time. The method was evaluated on 6 enrolled subjects after
obtaining all ethical approvals. The evaluations were conducted against two existing contact based methods;
thoracic and abdominal bands. Results showed a correlation coefficient of 0.9974 to 0.9999 depending on the
location of the ROI relative to the detected tip of the nose. The main contributions of the work was the
successful development and evaluation of the facial features tracking algorithms in thermal imagining, the
evaluation of thermal imaging as a technology for respiration monitoring in a hospital environment against
existing respiration monitoring systems as well as the real time nature of the method where the frame processing
time was 40 ms from capture to respiration feature plotting.
Keywords: - Image processing, Noncontact, Real-time, Respiration monitoring, Thermal imaging,
Thresholding.
INTRODUCTION
Oxygen is a naturally occurring element in the atmosphere which the body uses as part of a process to
obtain energy. Air-rich oxygen enters the lungs through inspiration from where oxygen goes into the blood
stream via the capillaries and is transported all over the body. During expiration de-oxygenated air containing
the produced carbon dioxide is returned to the atmosphere.
The average number of breathes taken over a predefined period of time is respiration rate. Its value
depends on several factors that include the person‟s age, gender and medical condition. It ranges between 12 to
60 cycles per minute in healthy adults and children. During respiration, a number of physiological effects take
place that are the bases for respiration monitoring. These include variations in air flow as air is inhaled and
exhaled, chest/abdomen movements, blood oxygen saturation variations, temperature changes in inhaled and
exhaled air and variations in the amount of emitted infrared radiation from the skin surface centred on the nose
and mouth.
An uneven respiration rate is a good indicator of imbalance of the normal body condition [1] that
studies [2-5] have shown to be a predictor of potential serious clinical events. Early detection and proactive
intervention of the predicted condition can avert a potential serious harm [6,7]. Respiration rate therefore, needs
to be accurately monitored to establish correct physiological operations in order to take the necessary clinical
steps. Existing gold standard devices for monitoring respiratory rate make contact with the subject‟s body
during monitoring and are therefore, referred to as contact based. They mostly require the subject to be
constrained and thus disturb the sleep position which affects normal breathing. Since a natural sleep position
does not alter breathing, it is therefore, critical that respiration monitoring does not alter it as well. In noncontact
respiration monitoring, the device is not attached to the subject‟s body and therefore, a more reliable medical
assessment can be obtained. Studies have reported that noncontact based methods for monitoring respiratory rate
are better than contact based methods [8-11]. They attributed this to improved patient‟s comfort, improved
accuracy due to elimination of distress as well as non-possibility of disruption of monitoring as a result of
dislodgement of the sensing device.
In this study, technique to facilitate noncontact respiration rate monitoring has been developed and its
effectiveness was evaluated. The technique was based on measuring detecting the amount of emitted infrared
radiation from the skin surface centred on the nose (thermal imaging method). This is based on a fact that during
respiration, exhaled air is warmer than inhaled air. The air temperature variation during respiration results in
temperature variation along the air flow path especially the area around the tip of the nose and upper lip regions
thereby resulting in corresponding variation in the amount of emitted infrared radiation from those areas. These
continuous variations in infrared emissions around the affected regions can be observed and processed by

Evaluation of Thresholding Based Noncontact Respiration Rate Monitoring using Thermal Imaging.
thermal imaging (thermography). A respiration signal can then be produced from the resulting images.
Respiration rate is then obtained by processing the resulting signals.
Methodology
1.1 Camera Setup
The emissivity of the thermal camera was set to 0.98 as required for skin temperature measurement.
The computer ran on Windows 7 operating system with a memory of 2.88 GB having an Intel Core processor
that ran at 2.6 GHz. The camera was controlled using software that was developed using the National
Instrument's LabVIEW environment. The camera's field of view (FOV) lens was 24 degrees and it was setup
such that the thermal image 𝑓(𝑥, 𝑦) acquired from it consisted mostly of the subject's head and shoulders.
Ethical approvals were obtained at Sheffield Hallam University, the Sheffield Children's Hospital and the
National Health Service (NHS), United Kingdom.
1.2 Experimental Setup
The camera was mounted on a tripod below the subject's upper body, at about 1 m from the subject
who sat comfortably on a chair in a research laboratory. The camera was inclined at about 20o
to focus upward
towards the subject's face such that the subject's upper body was contained in the camera's field of view. This
was to ensure that a good profile of the area most affected by respiration was adequately obtained. A typical
setup is shown in Error! Reference source not found..
Thoracic and abdominal piezoelectric bands were used to record respiratory signal from the subjects.
The mean temperature and relative humidity of the room were 24.4°C (standard deviation of 0.35) and 45%
(standard deviation of 0.71) respectively.
The frame capture rate per second was 20 in which allowed real time processing of the data and
provided a sufficient sample rate. Each subject was monitored for at least 5 minutes but typically more than 10
minutes. The camera was controlled using software that was developed using the National Instrument's
LabVIEW environment. The flow chart of the algorithm involved in this method is shown in Figure 0-1.
Acquire Image
Filter & Segment
Create ROI & extract
Respiration feature
Start
End
Reduce Image Resolution
Monitoring
terminated?
YES
NO
Locate tip of nose
Iterative
Elimination
Algorithm
Obtain Horizontal & Vertical
profiles of image
Detect face indicate tracking
Obtain maximum pixel
value on the face Pm
Threshold image
using (Pm-10) as
lower threshold
Iterative
Elimination
Algorithm
Locate eyes’ corners
Figure 0-1: Flow chart of feature detection based on thresholding algorithm

1.3 Face Detection
A facial region affected by respiration needed to be identified such that respiration induced temperature
changes could be reliably detected. The best feature that suited this purpose was the nose but due to varying
temperature of the skin area of the nose as a result of respiration posed consistency issues for the nose to be
directly located with accuracy in thermal imaging. In order avoid this inconsistency and detect the nose in
thermal image, a secondary feature needed to be located that would act as a reference to the nose. This would
also eliminate user interaction at any point during the recording and therefore eliminate possible errors that
could be introduced as a result. The two corners of the eyes met the requirement of being a reference in
detecting the nose for two reasons. Firstly, the corners of the eyes have the highest temperature on the face and
secondly the horn of the nose was always between the two corners of the eyes. The acquired image was
therefore segmented to remove objects with lower temperatures to that of the skin. Unwanted objects with
similar temperatures to that of the skin that could have been missed were then eliminated using an iterative
elimination algorithm developed and described in [12, 13]. Thereafter the face was detected and enclosed in a
rectangle to indicate tracking also described in [12]. The coordinates of the rectangle enclosing the face were
saved as:
rightBottomyx
leftBottomyx
rightTopyx
leftTopyx
br
bl
tr
tl




,
,
,
,
(0-1)
1.4 Eyes' Corners Detection
In [14] we detected the eyes‟ corners in a slightly different manner which came with some errors right
from detecting the face as well as with the eyes‟ corners detection. These observations were improved upon in
this evaluation by addition of equations 2-4 and 2-5 after taking the anthropometry of the face into consideration,
as will be explained shortly. Since the corners of the eyes had the highest temperature over the face, they also
had the highest pixel values by translation. To detect the corners of the eyes, the search for the maximum pixel
values was carried out within the bounding rectangle. For the first acquisition, the subject was required to face
the camera so that both eyes' corners could be detected. The reason for this would be explained later in the text.
Starting from 𝑥𝑙, 𝑦𝑡 the pixel value (i.e. the first pixel) was recorded as the maximum pixel. Its value and
coordinates were stored as 𝑃𝑚𝑎𝑥 and 𝑥 𝑝, 𝑦𝑝 respectively. The next pixel 𝑥𝑙+1, 𝑦𝑡 is looked at and compared with
the stored maximum, if it is higher, 𝑃𝑚𝑎𝑥 and 𝑥 𝑝, 𝑦𝑝 are replaced. This is repeated for each pixel value within the
bounded face. Upon reaching 𝑥 𝑟, 𝑦 𝑏 (i.e. the last bottom most pixel value within the bounded facial image), the
maximum pixel value 𝑃𝑚𝑎𝑥 and its corresponding coordinates 𝑥 𝑝, 𝑦𝑝 gives an indication of the corner of one eye.
One corner is always located first as the two eye corners do not have the same temperature, thus, the
highest pixel value on the other eye corner would be different from that already detected. However, there is the
likelihood that the next lower pixel value than 𝑃𝑚𝑎𝑥 was located next to 𝑃𝑚𝑎𝑥 and therefore still belong to the
detected eye corner. If that was the case, then, locating two maximum values would not yield a pointer to the
location of the two eyes‟ corners. To deal with these problems while avoiding complex and computational
intensive algorithms, the face in image was segmented using the maximum pixel value 𝑃𝑚𝑎𝑥 . The segmentation
was such that the eye corners would be returned as blobs with a maximum length (radius) of 10 pixels.
Thresholding was used to achieve the segmentation using:
 


 
 LTHgif1
Otherwiseyxg 0, (0-2)
Where 10max  PLTH
This was a valid assumption as the corner of the second eye cannot be 10 pixel values less than the
detected corner of the first eye. After the segmentation, all regions with pixel values greater than 𝑇𝐻𝐿constituted
the new image. These included regions associated with both eyes' corners as well as the side on the nose and in
some cases the mouth region. When only the two regions associated with the corners of the eyes were present,
the process of identifying the two corners was straight forward. It was however noted that some of the anomalies
above were present, between or besides the location of the eyes' corners.

Only the upper half of the facial image was considered in the search for the corners of the eyes. This
was because the eyes are located between the vertex and the upper half of the face. This decision was not only
valid but also reduced the computation time by about half the normal time if the whole face was considered.
The segmented image was then dilated to connect regions that were close enough to the blobs. Holes
were then filled and then followed by eroding the image to return it to the original size before the dilation. The
numbers of particles in the image were counted and the maximum horizontal and vertical coordinates occupied
by each blob was returned. The coordinate of each particle contained in a blob was used to obtain the
approximate coordinates of the centre of gravity of each blob using:








Ni
i iy
N
gy
Ni
i ix
N
gx
1
1
1
1
(0-3)
Where 𝑥 𝑔, 𝑦𝑔 represent the average positions of the central points of the horizontal and vertical sections
in a blob and 𝑥𝑖, 𝑦𝑖 are the horizontal and vertical coordinate of a pixel in reference.
The blob containing the maximum pixel value 𝑃𝑚𝑎𝑥 was identified in reference to the distances
between its coordinates 𝑥 𝑝, 𝑦𝑝 and the centre of mass of each blob. The blob that had the least distance to 𝑃𝑚𝑎𝑥
was identified as the first eye corner blob and was used to locate the second eye corner. The blob representing
the second eye corner was identified using the following steps:
A blob was considered and its maximum vertical pixel length (considered as the diameter) was
obtained and then divided by 2 to get the radius, r.
The y-coordinate 𝑦𝑔 of the centre of gravity of the detected first eye 𝑒1 corner was then used to check
whether the blob in reference (𝑒 𝑥) met the criteria:
       regyxegyregy  11 (0-4)
If the blob met the criteria, then it was identified as the corner of the second eye and the coordinates of
its centre of gravity 𝑥 𝑔, 𝑦𝑔 𝑒2 were stored, otherwise the next blob (if any) was considered. The anthropometry
of the face gave an indication of the possible separation between the inner corners of the two eyes (i.e. the
intercanthal width) B as one fifth of the head breadth H [15, 16]. Therefore the horizontal separation between
the x-coordinate 𝑥 𝑔 of the centre of gravity of the detected eye corner 𝑒1 and the blob in reference (𝑒 𝑥) must
also meet the criteria:
56
lxrx
B
lxrx 


(0-5)
The limit in equation 2-5 was introduced as a lower boundary to avoid detecting false eye corner.
When there was about 45 degree sideways head movement in either direction from when the face was looking
straight into the camera, the two eyes' corners could still be detected. When there was a considerable sideways
head movement however, only one eye corner was visible due to occlusion. Obtaining just one corner of an eye
was not enough to identify where the nose was located using this method. To deal with this situation where one
corner was not visible, the pixel distance between the two eyes' corners were calculated during the first iteration.
This was the reason for the requirement of the subject to face the camera at the beginning of acquisition, as
mentioned. The coordinates of both eyes' corners in each image were also saved and were replaced as new
locations were obtained. The distance between the two eyes' corners for each image was then compared to the
present distance and the x-coordinates 𝑥 𝑔(𝑒1) and 𝑥 𝑔(𝑒2) of both eyes' corners to those obtained (and saved)
during the previous iteration. A difference less than zero indicated that there was a movement to the left by the
subject (right in the image) and for a difference greater than zero, there was a movement to the right by the
subject (left in the image). This was so that the direction of head movement was known in order to identify
which eye corner was not in the camera's FOV.
Identifying the direction of eyes' corners movement was the reason why the subject was required to
face the camera for the first image acquisition. Obstruction of one eye corner could arise when the subject
moved their head in a sideways direction. Knowing the eye corner that was obstructed from the camera's view
allowed obtaining a coordinate location that would serve in detecting the location of the nose, just as the other
eye corner would have were it not obstructed. To obtain the appropriate coordinate when one eye corner was not
visible (due to obstruction) when the direction of the head movement was known as well as the maximum pixel
distance between the centre of gravity of the two eyes' corners, the rectangle that contained the face was used.
When there was a movement towards the left of the image that resulted in obstruction of the second eye corner

(i.e. the left eye corner was obstructed), then the coordinate of 𝑥𝑙 and 𝑦𝑔 𝑒1 were used as the appropriate
coordinate in detecting the nose. If on the other hand, the right corner of the eye were obstructed, then 𝑥 𝑟 and
𝑦𝑔 𝑒1 were used. In either case, the appropriate coordinates were labelled as 𝑒2 while those for the visible eye
corner detected labelled as 𝑒1. This was valid since only either the right or left sides of the nose was visible
when the left or right eye corner was obstructed respectively. There was no facial image beyond the left or the
right side of the rectangle bounding the face at those moments respectively.
1.5 Nose Detection
Looking at the face, the horn of the nose which is a part of the nose was always located between the
corners of the eyes. The nostril being the opening through which inhalation and exhalation take place during
nasal breathing was the lower end of the nose. With the location of the two eyes' corners detected, all that was
required was to identify the location of the nose. To do that, the image bounded by the coordinates of the eyes'
corners (𝑥 𝑔, 𝑦𝑔(𝑒1) and 𝑥 𝑔, 𝑦𝑔(𝑒2)) and that by the y-coordinates of the lower rectangle bounding the face in
reference to the x-coordinates of the eyes' corners (𝑥 𝑔(𝑒1), 𝑦 𝑏and 𝑥 𝑔(𝑒2), 𝑦 𝑏) was considered. This area of the
face contained the nose and mouth regions. For each acquired frame, we considered the highest and lowest pixel
values 𝑁𝑝𝑚𝑎𝑥 and 𝑁𝑝𝑚𝑖𝑛 respectively. The image within the area specified was then segmented so as to remove
all areas that were not associated with the nose. To do this, the image was segmented using:
   UTHgLTHif1
Otherwise0yx,g

 (0-6)
Where  
min
minmax
2
pL
pp
U NTHand
NN
TH 


Considering the maximum and minimum pixel values for each acquired frame was in order to consider
the pixel values that were affected by respiration, i.e. during expiration, the infrared radiated around the nose
increases and so does the pixel values and the reverse occurs during inspiration. This was also noted by [17]. If
constant values were considered, then the nose could not have been segmented and subsequently could not be
detected in some frames. After the segmentation, only the area representing the nose remained in the image and
a rectangle was created to enclose it to indicate detection. The tip of the nose 𝑡 𝑁 in frontal (or near frontal) view
is the mid-point between the left and right bottom sides of the region enclosing the nose. For non-frontal view,
the respective side of the facial boundary was used as the tip of the nose.
1.6 Region of Interest (ROI)
The region between the tip of the nostrils and the vertex of the lip was always affected by respiration.
The region was close to the tip of the nose. Four regions of interest (ROI) were created in order to get different
respiration signals from different locations that were affected by respiration. The regions were:
1. ROI1: The average pixel value of the area within the rectangle bounding the detected nose area.
2. ROI2: A circle with radius of 5 pixels located beneath the left nostril. Its centre was located on the bottom
line of the rectangle bounding the detected nose and at a quarter of the width of the nose area from the left.
3. ROI3: A circle with radius of 5 pixels located beneath the right nostril. Its centre was located on the
bottom line of the rectangle bounding the detected nose and at a quarter of the width of the nose area from
the right.
4. ROI4: The non-zero pixel values within obtained in equation 3-4.
5.
Results And Discussion
After locating the corner of an eye and thresholding the image as described, the resulting image
contained areas associated with only the two corners of the eyes in most cases. Figure 0-1(b) shows the resulting
image after thresholding and Figure 0-1(c) shows the two eyes' corners after considering the upper half objects
in the facial image.
Figure 0-1: (a) Original image (b) the results of thresholding based on the temperature profile of first
eye's corner and (c) two objects (only) representing the two corners of the eyes.

In the instance indicated in Figure 0-1, locating the two corners of the eyes was by obtaining the
coordinates of the two corners. In some cases however, areas associated with the sides of the nose, the mouth
region as well as some irregular areas associated with skin marks (e.g. scars) also appeared in the image due to
the temperature similarities of the regions as shown in Figure 0-2.
Figure 0-2: (a) Original image (b) the results of thresholding based on the temperature profile of first
eye's corner and (c) more than two objects.
In these cases, locating the second eye corner was not as straight forward. This was because, there was
no exact prior knowledge indicating when more than two objects would be present as against two, the second
eye's algorithm had to be used. Objects appearing below the center of mass of the face were rejected as
explained earlier. The performance of the second eye's algorithm in detecting the eyes-corner pair is shown in
Figure 0-3, where even when more than two objects appeared in the image.
„
Figure 0-3: (a) Detection of eyes' corners in presence of more than two objects and (b) eye corners
mapped onto image.
By detecting one corner of an eye, the other eye's corner was detected by the second eye's algorithm
even when there were objects not related to the eye. The nose was then detected after obtaining the position of
the two eyes‟ corners. The effectiveness of the nose detection is shown in Figure 0-4.
Figure 0-4: Detection of the nose in different images.

The respiration signals were obtained from the four ROIs as previously described. The locations of the
ROIs are shown in Figure 0-5 for a subject.
Figure 0-5: (a) ROI1, (b) ROI2, (c) ROI3 and (d) ROI4.
ROI1 averages all the pixel values which includes parts of the corners of the eyes and is therefore
expected to have low pixel average due to the larger area and the temperature differences between the corners of
the eyes‟ and area around the nose contained within the ROI. ROI2 and ROI3 are circles (in red) of 5 pixels
radius beneath the left and right nostrils respectively and should have high amplitudes as the temperature
distribution within the ROIs are almost uniform.
The respiration signal simultaneously obtained from the 4 ROIs in one of the recordings is shown in
Figure 0-6.
Figure 0-6: Respiration signals from the four ROIs
All the respiration signals were normalized around zero in order to enable overlay comparison. ROI2
and ROI3 have similar respiration waveform characteristics which are attributed to the location of the ROIs.
ROI1 have higher amplitude that ROI2 and ROI3 as it had a larger range between its minimum and maximum
average pixel values which was derived due to inspiration-expiration temperature difference. ROI4 had the
highest amplitude as it had the largest range that was attributed to its location (i.e. the nose). During inspiration,
the nose cools especially around the post inspiration pause period and during expiration it is warmed. The area
occupied by the nose in ROI4 is considerably reduced during inspiration and the pixel values are generally small
due to temperature drop. ROI4 is also inverted as due to the effect of averaging the pixels. During inspiration the
numbers of pixels are less and the temperature range is less giving a higher average than during expiration
0 10 20 30 40 50 60
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
t, sec.
Amplitude
ROI1
ROI2
ROI3
ROI4

where the temperature range is large and the numbers of pixels are also more thus giving a lower average which
the normalization ranges around the zero axes.
Figure 0-7 and Figure 0-8 show a minute-respiration signal of a subject comparing the threshold based
method obtained from ROI1 and ROI2 respectively against two ground truths (thoracic and abdominal bands)
derived.
Figure 0-7: The ROI1 derived against the ground truths respiration signal
The threshold derived respiration signal always lags behind the ground truths during inspiration while
during expiration it mostly lags and leads in few instances.
Figure 0-8: The ROI2 derived against the ground truths respiration signal
Table 0-1 provides a summary of the respiration rates from the 6 subjects included in the study
alongside that from the ground truths (i.e. the thoracic and abdominal bands). The mean age of the enrolled
subjects was 38.7 years with a standard deviation of 3.38 years.
Table 0-1: Measured respiration rates.
Subject Gender Age
Thermal Imaging Based Thoracic
Band
Based
Abdominal
Band
Based
ROI1 ROI2 ROI3 ROI4
FT1 Male 41 23.6 23.6 25.0 23.6 25.0 25.0
FT2 Male 41 18.9 18.9 18.9 18.9 19.0 19.0
FT3 Female 33 8.0 8.0 8.0 7.5 8.0 8.0
FT4 Male 36 9.6 9.6 9.6 9.6 9.7 9.7
FT5 Female 40 9.2 9.2 9.2 9.2 9.2 9.2
0 10 20 30 40 50 60
-0.4
-0.2
0
0.2
0.4
0.6
t, sec.
Amplitude
Abdominal Band Based
Thoracic Band Based
Thermal Imaging Based
0 10 20 30 40 50 60
-0.4
-0.2
0
0.2
0.4
0.6
t, sec.
Amplitude
Abdominal Band Based
Thoracic Band Based
Thermal Imaging Based

FT6 Male 41 16.9 16.9 16.9 16.9 16.9 16.9
For FT1, the ground truth RR was 25 cycles per minute in both, however, in thermal imaging threshold
based method the RR was 23.6 cycles per minute in ROI1, ROI2 and ROI4 derived respirations. This is a 1.4
cycles per minute less than the ground truth. Although ROI3 derived was the same as the ground truth, three of
the regions had less RR. FT2 and FT4 had RRs that were 0.1 cycles less that the corresponding ground truth
RRs in all the ROIs. FT5 and FT6 had the same RRs in all the ROIs as the ground truth. FT3 had the same RR
in all the ROIs as the ground truth except ROI4.
CONCLUSION
Thermal imaging based respiration rate monitoring based on threshold was developed. The method
detected the face of the subject then the two corners of the eyes and thereafter located the nose in thermal
images. Respiration features were extracted within created ROIs using a statistical measure. Respiration signals
were then obtained from succeeding features before obtaining the respiratory rate over a period of time. The
method was evaluated on 6 adult volunteers in a hospital setting as presented in the results and discussions
section.Six subjects participated in the evaluation of the respiration rate monitor based on thresholding. Four
ROIs were created around and on the detected nose region and respiration signals obtained simultaneously from
each. Correlation of the respiratory rates with the corresponding ground truths was at least 0.9977 in all the four
ROIs; ROI4 had 0.9977, ROI1 and ROI2 had 0.9984 while ROI3 had 0.9999.
ACKNOWLEDGEMENT
The author wishes to acknowledge the subjects that volunteered in the research and most especially his
research supervisory team and the sleep unit staff of the Sheffield Children Hospital.
REFERENCES
[1]. M. A. Cretikos, R. Bellomo, K. Hillman, J. Chen, S. Finfer and A. Flabouris, “Respiratory rate: the
neglected vital sign,” The Medical Journal of Australia, 188(11), 2008, pp. 657 - 659.
[2]. C. P. Subbe, R. G. Davies, E. Williams, P. Rutherford and L. Gemmell, “Effect of Introducing the
Modified Early Warning score on clinical outcomes, cardio-pulmonary arrests and intensive care
utilisation in acute medical admission.,” Anaesthesia, 58, 2003, 797-802.
[3]. Y. Nam, J. Lee and K. H. Chon, “Respiratory rate estimation from the built-in cameras of smartphones
and tablets.,” Annals of Biomedical Engineering, 42(4), 2014, 885-898.
[4]. F. Q. Al-Khalidi, “Facial tracking method for noncontact respiration rate monitoring,” Proc. 7th
International Symposium on Communication Systems Networks and Digital Signal Processing,
Newcastle upon Tyne, United Kingdom, 2010, 751-754
[5]. M. Cullinane, G. Findlay, C. Hargraves and S. Lucas, “"An acute problem?" National Confidential
Enquiry into Patient Outcome and Death.,” London, 2005.
[6]. D. Konrad, G. Jäderling, M. Bell, F. Granath, A. Ekbom and C. Martling, “Reducing in-hospital cardiac
arrests and hospital mortality by introducing a medical emergency team.,” Intensive Care Medicine ,
36(1), 2010, 100-106.
[7]. J. Hosking, J. Considine and N. Sands, “Recognising clinical deterioration in emergency department
patients.,” Australasian Emergency Nursing Journal, 17(5), 2014, 59-67.
[8]. R. Murthy, I. Pavlidis and P. Tsiamyrtzis, “Touchless monitoring of breathing function,” Proc. 26th
Annual International Conference of the IEEE Engineering in Medicine and Biology Society, California,
United States of America, 2004, 1195-1199.
[9]. F. Q. Al-Khalidi, R. Saatchi, D. Burke and H. Elphick, “Respiration rate monitoring: A review,” Pediatric
Pulmonol., 46, 2011, 523-529.
[10]. M. Bartula, T. Tigges and J. Muehlsteff, “Camera-based system for contactless monitoring of respiration,”
Proc. 35th Annual International Conference of the IEEE Engineering in Medicine and Biology Society,
Osaka, Japan, 2672-2675.
[11]. I. Smith, D. Krucheck, J. Mackay and N. Fahrid, “Respiratory rate measurement: A comparison of
methods.,” British Journal of Healthcare Assistants , 5(1), 2011, 18-23.
[12]. A. H. Alkali, R. Saatchi, H. Elphick, D. Burke, “Facial tracking in thermal images for real-time
noncontact respiration rate monitoring,” Proc. 2013 European Modelling Symposium, Manchester,
United Kingdom, 2013, 265-270.
[13]. A. H. Alkali, “Iterative Elimination Algorithm for Thermal Image Processing,” Arid Zone Journal of
Engineering, Technology and Environment, 10, 2014, 85-93.
[14]. A. H. Alkali, R. Saatchi, H. Elphick and D. Burke, “Eyes' corners detection in infrared images for real-
time noncontact respiration rate monitoring,” Proc.2014 World Congress on Computer Applications and
Information Systems, Hammamet, Tunisia, 2014, 1-5

[15]. J. W. Young, “Head and Face Anthropometry of Adult U.S. Citizens,” Federal Aviation Administration,
Wasington, DC, United States of America, 1993.
[16]. L. G. Farkas, M. J. Katic and C. R. Forrest, “International Anthropometric Study of Facial Morphology in
Various Ethnic Groups/Races.,” Journal of Craniofacial Surgery, 16, 2015, 615-646.
[17]. J. Fei and I. Pavlidis, “Thermistor at a Distance: Unobtrusive Measurement of Breathing.,” IEEE
Transactions on Biomedical Engineering, vol. 57, pp. 988 - 998, 2010.
*Abdulkadir Hamidu Alkali. "Evaluation of Thresholding Based Noncontact Respiration Rate
Monitoring using Thermal Imaging." International Refereed Journal of Engineering and Science
(IRJES) 6.7 (2017): 50-58.

Evaluation of Thresholding Based Noncontact Respiration Rate Monitoring using Thermal Imaging.

More Related Content

Similar to Evaluation of Thresholding Based Noncontact Respiration Rate Monitoring using Thermal Imaging. (20)

More from IRJESJOURNAL (20)

Recently uploaded (20)

Evaluation of Thresholding Based Noncontact Respiration Rate Monitoring using Thermal Imaging.