Contactless Optical Respiration Rate Measurement for a Fast Triage of

SARS-CoV-2 Patients in Hospitals

Carolin Wuerich

, Felix Wichum

, Christian Wiede

and Anton Grabmaier

Fraunhofer IMS, Finkenstrasse 61, Duisburg, Germany

Keywords:

Respiration Rate, Image Processing, Remote Measuring, SARS-CoV-2 Triage, Feature Tracking.

Abstract:

Especially in hospital entrances, it is important to spatially separate potentially SARS-CoV-2 infected patients

from other people to avoid further spreading of the disease. Whereas the evaluation of conventional laboratory

tests takes too long, the main symptoms, fever and shortness of breath, can indicate the presence of a SARS-

CoV-2 infection and can thus be considered for triage. Fever can be measured contactlessly using an infrared

sensor, but there are currently no systems for measuring the respiration rate in a similarly fast and contactless

way. Therefore, we propose an RGB-camera-based method to remotely determine the respiration rate for the

triage in hospitals. We detect and track image features on the thorax, band-pass ﬁlter the trajectories and

further reduce noise and artefacts by applying a principal component analysis. Finally, the respiration rate

is computed using Welch’s power spectral density estimate. Our contactless approach is focused on a fast

measurement and computation. It is especially adapted to the use case of the triage in hospitals by comprising

a face detection which is robust against partial occlusion allowing the patients to wear face masks. Moreover,

we show that our method is able to correctly determine the respiration frequency for standing patients despite

considerable body sway.

1 INTRODUCTION

The entire world is currently experiencing an un-

precedented pandemic with COVID-19. As more

and more people become infected with SARS-CoV-

2, hospital capacities are also becoming scarcer. It

is important to effectively protect patients, medical

staff and visitors from potentially SARS-CoV-2 in-

fected patients. Hereby, the most efﬁcient solution

is a spatial separation. However, for the separa-

tion, potentially infected patients need to be identiﬁed

quickly, and conventional laboratory tests based on

polymerase chain reaction (PCR), antibodies or anti-

gens take too long to evaluate. In addition, if there

is a high incidence of infection, not everyone can be

tested due to limited laboratory capacity.

Since laboratory diagnoses are not possible or

practically feasible, a focus on the symptoms of

SARS-CoV-2 is necessary. The two most com-

mon symptoms are fever and accelerated breathing.

https://orcid.org/0000-0003-0917-2696

https://orcid.org/0000-0002-3586-2802

https://orcid.org/0000-0002-2511-4659

https://orcid.org/0000-0002-4882-4223

Whereas the ﬁrst one can be measured quickly and

contactlessly with the help of a thermal imaging cam-

era, there is no contactless measuring system for de-

termining the respiration rate which can take mea-

surements quickly and in an uncomplicated way.

In order to provide a solution, we propose an op-

tical measuring system which can automatically de-

termine the respiration rate. Such measuring systems

have already been used in the ﬁeld of Ambient As-

sisted Living (Wiede et al., 2019). The principle is

based on tracking the optical ﬂow of suitable features

on the thorax, ﬁltering of the trajectories with band-

pass ﬁlters, artefact reduction by principal component

analysis (PCA) and frequency analysis. Since the ap-

plication scenarios differ signiﬁcantly, it is necessary

to investigate which changes need to be made for the

new ﬁeld of application. This concerns questions of

save distancing, face detection with masks and body

movement during standing, and affects illumination,

camera attachment and especially the algorithm de-

sign.

The system must enable fast measurements, in or-

der to separate the patients as quickly as possible at

the hospital entrance. A great advantage of the pro-

posed measurement method is that there is no contact

Wuerich, C., Wichum, F., Wiede, C. and Grabmaier, A.

Contactless Optical Respiration Rate Measurement for a Fast Triage of SARS-CoV-2 Patients in Hospitals.

DOI: 10.5220/0010408400290035

In Proceedings of the International Conference on Image Processing and Vision Engineering (IMPROVE 2021), pages 29-35

ISBN: 978-989-758-511-1

between the patient to be measured and the measure-

ment system, which eliminates the risk of infection

carriers being passed on.

In this paper, we ﬁrstly present the state of the

art in remote respiration rate determination in Section

2. In Section 3, we explain the used methods in de-

tail. This is followed by the experiments and results

in Section 4. The discussion is presented in Section

5. We conclude with our ﬁndings and the outlook on

future developments in Section 6.

2 STATE OF THE ART

In clinical practice, there are multiple methods to

measure the respiration rate such as nasal thermistors,

pressure transducer, respiratory effort belts or an anal-

ysis of the electrocardiogram (ECG). However, these

devices all require direct contact to the patient. This

increases the risk of contamination of the device and

an infection of the nurse who needs to attach the de-

vices to the patient. As alternative, optical methods

can contribute to the protection of medical staff and

other patients by enabling a safety distance.

Jin Fei and Pavlidis (2010) developed a remote

thermistor by applying a face detection followed by

a nostrils detection on thermal images. Around the

nostrils, wavelets are used to analyse changes in tem-

perature during inhalation and exhalation. Martinez

and Stiefelhagen (2012) use a Kinect camera to track

an infrared pattern projected on the upper body of

sleeping persons. Then, they deploy a Principal Com-

ponent Analysis (PCA) merging the trajectories and

an Auto-regressive (AR) spectral analysis to compute

the respiration frequency. The method by Lim et al.

(2014) uses a Kinect camera as well, but their ap-

proach is based on a moving average ﬁlter and a spline

interpolation on the depth points. Ostadabbas et al.

(2015) extended this approach by an automatic selec-

tion of a Region of Interest (ROI) on the chest.

While these approaches are based on the infrared

spectrum, we want to focus on applications of con-

ventional RGB-cameras in the visible light spectrum,

since these cameras are more economically priced

and thus can provide a low-cost system with a higher

availability. Using an RGB-camera, Tan et al. (2010)

observe the chest movement by subtracting two con-

secutive frames and analysing the generated edges.

Bartula et al. (2013) transform the images or selected

ROIs to 1D-vectors and then cross-correlate con-

secutive frames to obtain the motion-based changes.

Reyes et al. (2016) analyse the intensity changes in

the video channels caused by lifting and lowering the

chest during breathing. Massaroni et al. (2018) follow

a similar approach considering the intensity changes

of the video, but they select the ROI at the pit of the

neck.

An alternative approach considers the intensity

changes of the skin to determine the heart rate. From

a modulation of this heart rate signal, the respiration

rate can be derived (Poh et al., 2011; Tarassenko et al.,

2014). Poh et al. (2011) apply an independent compo-

nent analysis (ICA) for the frequency determination,

whereas Tarassenko et al. (2014) use AR models.

Further methods employ feature tracking and op-

tical ﬂow, such as Lukac et al. (2014) who use a KLT

tracker (Tomasi and Kanade, 1991), or Koolen et al.

(2015) who extended this approach by applying a

PCA, an ICA and ﬁnally a Short Time Fourier Trans-

form (STFT) to improve signal analysis. Wiede et al.

(2017) apply feature detection and tracking within a

selected ROI, and then use a PCA and FFT to deter-

mine the frequency.

Previous methods are not developed and tested for

the application of epidemic triage in hospitals. We

base our approach on the method by Wiede et al.

(2017) and adapt the algorithm to the triage use case

in hospitals. We reduce the total recording and com-

putation time to avoid accumulations of patients wait-

ing. Further, we improve the face detection by in-

creasing the robustness against occlusion to allow a

respiration rate measurement with face masks. More-

over, patients should touch as few objects and sur-

faces as possible making a respiration measurement

while standing desirable. Therefore, we show, that

our improved method is able to determine the respi-

ration rate correctly despite the body’s sway during

standing.

3 METHOD

3.1 System Overview

The proposed system allows the contactless measure-

ment of the respiration rate. For this purpose, an

RGB-camera is located 1.5 m away from the subject.

The use of the system in an application-like environ-

ment is shown in Figure 1.

In the following we use the term real-time respira-

tion rate (RTRR) to describe our proposed method. In

parallel to the image acquisition by an RGB-camera,

the face or alternatively the upper body of the patient

is detected in the video. Based on these results, the

ROI is placed on the chest and the area is further di-

vided into four subregions. Prominent features are

found and tracked over time. Hereby, the motion in

y-direction corresponds to the breathing movement.

IMPROVE 2021 - International Conference on Image Processing and Vision Engineering

Figure 1: The proposed system for contactlessly measur-

ing the respiratory rate in a hospital-like environment. The

camera can be located 1.5 m away from the subject and the

evaluation can be done in a safe distance or behind a glass

panel by medical personnel.

This is followed by a PCA and bandpass ﬁltering for

the removal of artefacts and noise. The respiratory

rate is determined by the highest spectral density of

the principal components. The average respiratory

rate is determined over 30 s time intervals. Figure

2 shows the different steps of the proposed method.

3.2 ROI Selection and Feature Tracking

To determine the respiration rate, the proposed ap-

proach observes the lifting and lowering movements

of the torso. For this, an ROI on the chest is to be

selected. Since the upper body detection is less ac-

curate, we ﬁrst employ a face detection algorithm to

estimate the ROI based on the relative position and

size of the bounding box of the face. Only if the face

is not visible, the upper body detection is applied.

First, we apply the Histogram of oriented Gradi-

ents (HoG) method (Dalal and Triggs, 2005) for face

detection. The employed model is built of ﬁve HoG

ﬁlters for different angles of the face. It is a rather

light-weight model and works well for frontal faces

and faces under partial occlusion. Latter is impor-

tant for our application since the patients might be

wearing masks. If this method does not ﬁnd a face

in the image, a Viola-Jones face detector (Viola and

Jones, 2004) is applied as alternative. Based on this,

the ROI is placed centrally below the bounding box

shifted and sized relatively to the bounding box of the

face. This way, we ensure the right size of the ROI

independently of the patient’s size or distance to the

camera. The factors for determining the relative po-

Figure 2: Overview of the proposed method for contactless

real-time measurement of the respiration rate. The image

processing includes the ROI selection and feature tracking,

and is followed by the time signal processing with the signal

extraction and frequency analysis.

sition of the ROI (see Equation 1a to 2d) were found

empirically. The ROI is deﬁned by the position of the

upper left corner (x

ROI

, y

ROI

) and its width w

ROI

and

height h

ROI

. Equivalently, the bounding box is deﬁned

by the position of the upper left corner (x

f ace

, y

f ace

)

and its width w

f ace

and height h

f ace

ROI

= x

f ace

− (0.7 · w

f ace

) (1a)

ROI

= y

f ace

− (1.4 · h

f ace

) (1b)

ROI

= 0.7 · h

f ace

(1c)

ROI

= 1.4 · w

f ace

(1d)

If non of the face detection algorithms was suc-

cessful, an upper body Viola-Jones detector is de-

ployed. The ROI is then to be selected relatively to

the bounding box of the torso, which is deﬁned by the

position of the upper left corner (x

torso

, y

torso

) and its

width w

torso

and height h

torso

ROI

= x

torso

+ (0.32 · w

torso

) (2a)

ROI

= y

torso

− (0.16 · h

torso

) (2b)

ROI

= 0.2 · h

torso

(2c)

ROI

= 0.36 · w

torso

(2d)

After computing the position of the ROI, it is split

equally into four subregions in order to improve the

Contactless Optical Respiration Rate Measurement for a Fast Triage of SARS-CoV-2 Patients in Hospitals

Figure 3: The face is detected to set the position of the

ROI relatively to the position and size of the face in the im-

age. Subsequently the ROI is divided into four subregions

in which the strongest features are identiﬁed and tracked.

robustness of the following signal processing. Within

each of these four subregions, we identify 15 unique

image features to be tracked subsequently. For this

purpose, we select the 15 strongest minimum Eigen-

value features according to the Shi-Tomasi feature de-

tection (Shi and Tomasi, 1993). Figure 3 shows a sin-

gle frame with the bounding box around the detected

face, the partitioned ROI and its selected features.

To observe the movement of the torso, the KLT

tracker by Tomasi and Kanade (1991) is applied to the

previously selected features. According to the bright-

ness assumption, the intensity of a pixel remains

constant for small movements and a short time pe-

riod. However, to enable tracking in two-dimensional

space, Tomasi and Kanade (1991) suggested to con-

sider the neighbouring pixels as well, resulting in a

3x3 patch to be tracked.

3.3 Time Signal Processing

In the further processing, only the y-direction of

the tracked features are considered, since this corre-

sponds to the main direction of the chest movement

during breathing. For each of the subregions, the y-

directories of the tracked feature points are averaged

and the mean y-displacement Y

avg

of all features is

considered to represent the overall movement of the

subregion, resulting in four time signals, see Equa-

tion 3. The indices i refer to the four regions with n

tracked feature points. If in any subregion all features

are lost during tracking or no features were found at

all, its ﬁnal signal is replaced by the signal of the sub-

region with the most features in the end.

avg

(t) =

∑

j=1

(t) (3)

Subsequently, a bandpass BP ﬁlter is applied to

exclude frequencies not in the range of the physio-

logical human respiration rate, see Equation 4. In or-

der to guarantee a linear phase response and therefore

a constant group delay, an FIR ﬁlter with 128 ﬁlter

coefﬁcients was chosen. The cutoff frequencies al-

low frequencies between 0.125 Hz (7.5 bpm) and 0.7

Hz (42 bpm) to pass, where bpm denotes breaths per

minute. This bandwidth complies with the physiolog-

ical human respiration rate.

(t) = Y

avg

(t) ∗ BP(t) (4)

In the next step the signal is mean-centred, see

Equation 5. In this context Y stands for the tempo-

ral average of Y

avg

. Afterwards the PCA is applied.

By splitting the signal in its principal components PC,

artefacts and noise are further reduced.

centered

(t) = Y

(t) −Y

(5)

For the ﬁnal frequency determination, the spectral

estimator described by Welch (1967) is employed us-

ing a sliding window of 30 seconds. It is applied to

each of the principal components, and ﬁnally, for the

determination of the respiration, the principal compo-

nent exhibiting the highest overall spectral density is

chosen. The respiration rate is calculated according

to Equation 6.

respiration

= max(|Welch(PC)|) (6)

The conversion of the respiratory frequency into the

unit bpm is done by a ﬁnal multiplication with the

factor 60.

4 EXPERIMENTS AND RESULTS

4.1 Experimental Setup

The validation of the approach is divided into two

parts. The ﬁrst set of experiments aims at testing

the general functionality of the implemented method,

whereas the second set of experiments focuses on the

applicability for the triage use case in particular.

In a ﬁrst experiment, we recorded a total of 24

videos of eight different persons. The videos to be

analysed are captured by an Allied Manta G-201-

RGB-camera at a ﬁxed frame rate of 10 fps and have

a resolution of 1624 x 1234 pixels. Each video has

IMPROVE 2021 - International Conference on Image Processing and Vision Engineering

a duration of 30 seconds, which is the total record-

ing time needed for the calculation of the respira-

tion rate by the algorithm. During this time the sub-

jects were sitting in a chair facing the camera and

were instructed not to talk. The experiments in-

cluded three female and ﬁve male subjects between

the age of 22 and 28, where each of the subjects was

recorded three times at different respiration rates. The

clothes of the subjects include plain as well as pat-

terned/textured shirts. This also includes very dark

and bright colours, which might be more challeng-

ing for the image processing algorithms. Light con-

ditions, exposure time and white balancing were not

changed during the experiments. Our method is im-

plemented on a NUC8i5BEK with an Intel

 Core

i5 processor (8

Gen).

Simultaneously to the recording of the video, we

measured the respiration with a reference system to

obtain ground truth data for validation. The reference

system consisted of a NeXus-10 MKII device with

the corresponding chest strap and the software Bio

Trace+ (V 2017 A). The resolution of the measured

respiration rate values is 0.1 bpm.

In a second experiment, we tested the validity of

the approach for the triage use case. Since the pa-

tients should touch as few objects and surfaces as pos-

sible, a respiration measurement while standing is de-

sirable. Further, the face detection algorithm has to

prove to be robust against occlusion due to masks.

We included these two challenges in the second set of

tests by recording the subjects with masks and while

standing. Again, we took three videos and reference

measurements for each of the eight subjects with the

same technical settings as before.

4.2 Results

The presented optical method rounds the measure-

ment results to whole numbers, while the data from

the reference system is rounded to one decimal place.

The recorded respiration rates range from 7.3 bpm to

37.0 bpm. The maximum absolute error is 3.2 bpm,

while the root mean squared error (RMSE) is 1.4 bpm

and thus within the self requested level of 3 bpm. For

the data with face mask and standing the RMSE is

1.3 bpm and without mask 1.5 bpm respectively. The

absolute errors of each individual measurement are

shown in Table 1. Figure 4 shows the error for each

target value. The two largest errors are located at the

outer edges of the measuring rate. Only with subject

3 measurement 3 the error exceeds the value of 3.0

bpm. Most absolute errors are even below 2.0 bpm

and a large proportion fall below an absolute error of

1.0 bpm.

Figure 5 shows the distribution of errors. There is

only one peak in the error probability function at -0.7

bmp . Starting from this point, the probability of error

occurrence decreases continuously in both directions

symmetrically. The measurements with face mask

and standing lie within the other results and show a

smaller spread.

A complete measurement takes 32.83 s on aver-

age in the C++ implementation. Thereof, 30 seconds

are required for the video recording (and parallel im-

age processing) and 2.83 s for the actual time signal

processing.

5 DISCUSSION

As shown in the previous section, the RMSE on our

data sets is at 1.5 bpm for sitting subjects without

mask and 1.3 bpm for standing subjects with mask.

Under the assumption that a healthy person has a res-

Table 1: Measurement results for sixteen subjects with three

measurements each. The ﬁrst eight subjects were sitting and

did not wear any face masks. The subjects nine to sixteen

were standing and did wear a face mask. The values of

the Nexus system (reference) and the RTRR approach are

shown.

Subject Reference [bpm] RTRR [bpm]

#1 #2 #3 #1 #2 #3

1 7.3 8.0 14.0 7 11 14

2 13.0 12.0 19.7 14 11 21

3 16.0 19.2 33.2 14 18 30

4 10.2 13.6 23.8 9 11 23

5 10.2 9.6 11.8 11 9 11

6 20.0 19.4 24.4 18 21 23

7 9.5 14.2 29.0 9 14 28

8 13.0 10.6 14.0 11 11 14

9 18.6 15.6 20.0 18 14 18

10 11.9 19.4 19.0 11 18 18

11 21.0 32.0 32.0 21 32 30

12 11.8 26.8 37.0 11 25 37

13 12.0 12.2 25.0 11 14 23

14 15.6 9.2 28.2 16 7 28

15 10.6 12.0 11.6 9 11 11

16 9.4 11.0 37.4 9 9 37

Contactless Optical Respiration Rate Measurement for a Fast Triage of SARS-CoV-2 Patients in Hospitals

10 20 30 40

Reference value [bpm]

Absolute measurement error [bpm]

without face mask

with face mask

Figure 4: Occurred absolute errors for our proposed RTRR

method in relation to the reference system.

Without face mask With face mask

−3

−2

−1

Measurement error [bpm]

Figure 5: Distribution of measurement errors.

piration rate between 10 and 15 bpm, the relative er-

ror would be 9 % to 14 %. This is acceptable for the

use case of the triage. It does not matter whether the

respiration rate is at 10 or 11 bpm respectively but

it should be distinguished between normal respiration

and an abnormal high respiration rate above 20 bpm.

This requirements can be met.

The reasons for the deviations are manifold. On

the one hand, it must be noted that the lighting con-

ditions were not always ideal. This results in partly

very dark or overexposed images, which do not allow

robust tracking of the features. As a solution in the

future, an ROI speciﬁc brightness control has to be

designed. It is also evident that the images of sitting

persons have a 0.2 bpm higher RMSE. This can be

explained by the fact, that the breathing patterns are

better visible while standing. In two of the tests with

masks, the subjects’ faces were not detected at ﬁrst.

After around 10 seconds, the faces were ﬁnally recog-

nised and the respiration rate could still be determined

without problems. The difﬁculties in face detection

can be attributed to signiﬁcant occlusion because of

the mask. The effect is intensiﬁed if subjects wear

glasses that reﬂect in the image, have bangs or poten-

tial additional face coverings. In these cases the face

is not sufﬁciently visible for robust recognition. A so-

lution could be provided by specially trained face de-

tectors or by asking the subjects to e.g. take off their

glasses. On the other hand, we showed that the im-

plemented method is robust against body sway during

standing. Although this movement often is stronger

than the breathing movement of the upper body, it is

not as constant in frequency. Therefore, Welch’s spec-

tral density estimation can effectively eliminate those

artefacts.

We are fully aware that the data set is rather small

for an extensive overview. Especially, the involve-

ment of different user groups of different age and pre-

existing conditions justify a following clinical study.

This clinical study will be carried out in the Univer-

sity Hospital Essen. Nevertheless, this paper focuses

on the technical part and we would like to share our

preliminary results with the community.

An important factor for a triage in the entrance

area of the hospital is time. A measurement should

take place as quickly as possible and at the same time

be as accurate as possible. The actual measurement is

performed in a 30 second measurement window, fol-

lowed by an evaluation with a processing time below

3 seconds. With the resulting total duration it is pos-

sible to examine more than 100 persons in one hour.

6 CONCLUSION

In this study, we presented a contactless optical ap-

proach to determine the respiration rate in an hospi-

tal entrance for triage. Thereby, the method is based

on an optical ﬂow, a PCA and a frequency analysis.

We could demonstrate that the results are accurate

(RMSE of 1.4 bpm) and fast (less than 35 seconds)

at the same time.

A clinical study to validate the results in a clini-

cal setting is already planned. Further work will con-

centrate on including further vital parameters such as

IMPROVE 2021 - International Conference on Image Processing and Vision Engineering

heart rate and blood pressure to the contactless mea-

surement procedure. Thereby, the focus should not

be solely on COVID-19 but as well on other diseases

with different symptoms. Furthermore, we intend

to expand our idea of a remote respiration rate de-

termination to other application areas such as access

control to buildings and public transport, assistance

robots and recovery monitoring.

REFERENCES

Bartula, M., Tigges, T., and Muehlsteff, J. (2013). Camera-

based system for contactless monitoring of respira-

tion. In Engineering in Medicine and Biology Society

(EMBC), 2013 35th Annual International Conference

of the IEEE, pages 2672–2675. IEEE.

Dalal, N. and Triggs, B. (2005). Histograms of oriented

gradients for human detection. In 2005 IEEE com-

puter society conference on computer vision and pat-

tern recognition (CVPR’05), volume 1, pages 886–

893. IEEE.

Jin Fei and Pavlidis, I. (2010). Thermistor at a Distance:

Unobtrusive Measurement of Breathing. IEEE Trans-

actions on Biomedical Engineering, 57(4):988–998.

Koolen, N., Decroupet, O., Dereymaeker, A., Jansen, K.,

Vervisch, J., Matic, V., Vanrumste, B., Naulaers, G.,

Van Huffel, S., and De Vos, M. (2015). Automated

Respiration Detection from Neonatal Video Data:. In

Proceedings of the International Conference on Pat-

tern Recognition Applications and Methods, pages

164–169. SCITEPRESS - Science and and Technol-

ogy Publications.

Lim, S. H., Golkar, E., Rahni, A., and Ashrani, A. (2014).

Respiratory motion tracking using the kinect camera.

In Biomedical Engineering and Sciences (IECBES),

2014 IEEE Conference on, pages 797–800. IEEE.

Lukac, T., Pucik, J., and Chrenko, L. (2014). Contactless

recognition of respiration phases using web camera.

In Radioelektronika (RADIOELEKTRONIKA), 2014

24th International Conference, pages 1–4. IEEE.

Martinez, M. and Stiefelhagen, R. (2012). Breath rate mon-

itoring during sleep using near-ir imagery and pca. In

Pattern Recognition (ICPR), 2012 21st International

Conference on, pages 3472–3475. IEEE.

Massaroni, C., Lopes, D. S., Lo Presti, D., Schena, E., and

Silvestri, S. (2018). Contactless monitoring of breath-

ing patterns and respiratory rate at the pit of the neck:

A single camera approach. Journal of Sensors, 2018.

Ostadabbas, S., Sebkhi, N., Zhang, M., Rahim, S., Ander-

son, L. J., Lee, F. E.-H., and Ghovanloo, M. (2015). A

Vision-Based Respiration Monitoring System for Pas-

sive Airway Resistance Estimation. IEEE Transac-

tions on Biomedical Engineering, pages 1–1.

Poh, M.-Z., McDuff, D., and Picard, R. (2011). Advance-

ments in Noncontact, Multiparameter Physiological

Measurements Using a Webcam. Biomedical Engi-

neering, IEEE Transactions on, 58(1):7–11.

Reyes, B. A., Reljin, N., Kong, Y., Nam, Y., and Chon,

K. H. (2016). Tidal volume and instantaneous respira-

tion rate estimation using a volumetric surrogate sig-

nal acquired via a smartphone camera. IEEE journal

of biomedical and health informatics, 21(3):764–777.

Shi, J. and Tomasi, C. (1993). Good Features to Track.

Technical report, Cornell University, Ithaca, NY,

USA.

Tan, K. S., Saatchi, R., Elphick, H., and Burke, D. (2010).

Real-time vision based respiration monitoring sys-

tem. In Communication Systems Networks and Digital

Signal Processing (CSNDSP), 2010 7th International

Symposium on, pages 770–774. IEEE.

Tarassenko, L., Villarroel, M., Guazzi, A., Jorge, J., Clifton,

D. A., and Pugh, C. (2014). Non-contact video-

based vital sign monitoring using ambient light and

auto-regressive models. Physiological Measurement,

35(5):807–831.

Tomasi, C. and Kanade, T. (1991). Detection and Tracking

of Point Features. Technical report, Carnegie Mellon

University.

Viola, P. and Jones, M. J. (2004). Robust real-time face

detection. International Journal of Computer Vision,

57(2):137–154.

Welch, P. (1967). The use of fast fourier transform for

the estimation of power spectra: a method based on

time averaging over short, modiﬁed periodograms.

IEEE Transactions on audio and electroacoustics,

15(2):70–73.

Wiede, C., Richter, J., and Hirtz, G. (2019). Contact-less vi-

tal parameter determination: An e-health solution for

elderly care. In VISIGRAPP (5: VISAPP), pages 908–

915.

Wiede, C., Richter, J., Manuel, M., and Hirtz, G. (2017).

Remote respiration rate determination in video data-

vital parameter extraction based on optical ﬂow and

principal component analysis. In International Con-

ference on Computer Vision Theory and Applications,

volume 5, pages 326–333. SCITEPRESS.

Contactless Optical Respiration Rate Measurement for a Fast Triage of SARS-CoV-2 Patients in Hospitals