Contact-less Vital Parameter Determination: An e-Health Solution for

Elderly Care

Christian Wiede, Julia Richter and Gangolf Hirtz

Department of Electrical Engineering and Information Technology,

Chemnitz University of Technology, Reichenhainer Str. 70, 09126 Chemnitz, Germany

Keywords:

Remote Vital Parameter Determination, Home Environments, e-Health, Contact-less, Heart Rate, Respiration

Rate, Oxygen Saturation, Blood Pressure.

Abstract:

Vital parameters are key ﬁgures for the basis functions of the human body. Without these basis body functions,

such as the heart beat, life is impossible. Therefore, vital parameters are indicators for a person’s general

medical condition. In recent years, the topic of vital parameter monitoring has been increasingly studied

in the ﬁeld of e-health. Especially the contact-less determination of vital parameters, such as heart rate,

respiration rate, oxygen saturation and blood pressure, with consumer cameras brings a variety of advantages.

In this work, we present methods to determine the mentioned vital parameters in a contact-less, optical way.

Furthermore, we evaluated these methods for an utilisation in home environments with respect to elderly care.

As a result, the remote determination of heart and respiration rate show reliable measurements, which makes

the proposed methods ready for the application in home environments.

1 INTRODUCTION

e-Health has become an evolving ﬁeld in recent ye-

ars. The business consultancy Roland Berger expects

that the e-health market will grow by 21 % every year

(Berger et al., 2016). Thereby, e-health is a collective

term for all applications of digital technology in the

health care sector. That comprises electronic health

records in the same way as telemedicine or consumer

health informatics. Especially self-monitoring health

care devices, such as ﬁtness trackers, mobile apps,

blood glucose monitors or blood pressure monitors,

have shown to become of increasing interest in our

society. The devices enable persons to analyse their

own health status in an independent way.

However, all of these consumer products require

body contact. This can be obstructive for persons

with sensible skin and bears discomfort in wearing.

Moreover, elderly people tend to forget to wear these

devices because of dementia or other cognitive disa-

bilities. In order to overcome these disadvantages, we

propose to extract the vital parameters heart rate, res-

piration rate, oxygen saturation and blood pressure re-

motely in home environments. This vital parameter

data can be recorded and analysed automatically. In

case of emergencies, an alarm can be triggered. More-

over, the continuous monitoring can be a key element

for an improved evidence-based diagnosis.

In this work, we want to outline the state of the

art of remote vital parameter determination and pre-

sent own methods to solve the research tasks. Hereby,

we do not only carry out experiments in laboratory

settings, but especially in realistic home environment

scenarios. The aim is to investigate whether the single

parameters can be measured accurately in the setting

of home environments.

2 REMOTE HEART RATE

DETERMINATION

2.1 Related Work

In the past, a human’s heart rate was obtained by con-

ventional methods such as the electrocardiogram or

the photoplethysmography (PPG), which was propo-

sed by Hertzman and Spealman (Hertzman and Spe-

alman, 1937) and is based on determining the volu-

metric changes in the tissue optically.

By extending the original transmissive approach

of the PPG, it is possible to determine the heart rate

by reﬂective light (Humphreys et al., 2005). Ver-

kruysse et al. continued with this idea and determi-

908

Wiede, C., Richter, J. and Hirtz, G.

Contact-less Vital Parameter Determination: An e-Health Solution for Elderly Care.

DOI: 10.5220/0007682809080915

In Proceedings of the 14th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2019), pages 908-915

ISBN: 978-989-758-354-4

ned the heart rate in the visible light spectrum by an

RGB-camera (Verkruysse et al., 2008). The ﬁrst au-

tomated approach in the visible light spectrum was

proposed by Poh et al. (Poh et al., 2010). Further im-

provements consist in using temporal ﬁlters (van Gas-

tel et al., 2014), autoregressive models (Tarassenko

et al., 2014) or adaptive ﬁltering (Wiede et al., 2016).

This group of methods are considered as so-called

intensity-based methods.

The other group of methods is called motion-

based methods, because it makes use of small mo-

vements of the head that are caused by the heart bump

induced blood ﬂow according to the the 3

Newto-

nian law. Balakrishnan et al. were the ﬁrst to apply

this principle (Balakrishnan et al., 2013) by tracking

this motion with optical ﬂow.

Nevertheless, all these methods do not eliminate

the underlying artefacts or are not applicable for home

environments.

2.2 Proposed Method

As outlined in the previous section, eliminating inten-

sity and motion artefacts is the key factor for determi-

ning the heart rate robustly.

In the ﬁrst step, a white balancing is necessary in

order to compensate the differences in image acquisi-

tion of different cameras. For our implementation, we

use the method presented by Garud et al. (Garud et al.,

2014). In the next step, the face is detected within

the image, because the face has both: A large amount

of skin pixels and a good detectability with standard

algorithms. We apply the face detector of Zhu and

Ramanan (Zhu and Ramanan, 2012), because it not

only provides the location of the image but as well

its orientation based on facial landmarks. Afterwards,

face regions with only skin, no hairs and less mo-

vements artefacts are selected, i. e. the forehead, the

nose and the two cheeks. By transferring all pixels of

these so-called regions of interest (ROI) to the HSV

colour space, it is possible to determine an individual

skin colour model for each person.

In the following, the video sequence can be conti-

nuously captured and the white balancing can be app-

lied to each single image. The face can be tracked by

a mixed approach of KLT tracking (Tomasi and Ka-

nade, 1991) and tracking-by-detection. Subsequently,

all pixels in the face matching the skin colour model

criteria are extracted for the further processing. These

extracted pixels are averaged in every time step for the

three colour channels red, green and blue.

The three colour channels are normalised and

bandpass ﬁltered in order to exclude implausible fre-

quencies. An independent component analysis (ICA)

is applied to split the three ﬁltered channels in one

wanted signal and two noise signals. The result are

three independent components. For each of them, the

periodicity is calculated to determine the wanted sig-

nal component. For this component, the dominant

frequency is computed by the Fast Fourier Transform

(FFT) within a sliding window. This dominant fre-

quency represents the heart rate. In order to prevent

this algorithm from measuring sudden changes in the

heart rate, an adaptive ﬁltering is implemented as low-

pass.

This algorithm shows an averaged root-mean-

square error (RMSE) ranging from 1.19 bumps per

minute (BPM) to 2.93 BPM for selected lab scenarios

(Wiede et al., 2018). These results have to be veriﬁed

under realistic conditions in home environments.

2.3 Adjustments for Home

Environment Measurements

Determining the heart rate in home environments

holds some challenges in comparison to the labora-

tory scenarios. Normally, the cameras are ﬁxed at

the ceiling in home environments, so that they do not

disturb the residents. On the one hand, a position at

the ceiling is advantageous because of less occurring

occlusion of objects and persons in the image. On the

other hand, this is accompanied by a perspective vie-

wpoint change, which leads to challenges in detecting

a human’s face. In Figure 1, such an example of per-

spective viewpoint change, with a very steep viewing

angle can been seen.

Moreover, the distance to the camera is higher

than in the lab scenarios. This goes hand in hand with

a smaller spatial resolution, which means that the face

is represented by less pixels. Furthermore, it cannot

be guaranteed that a person is always upright in an

image. Intensity artefacts strongly depend on the cer-

tain lighting in the room and can inﬂuence the mea-

surements. In addition to that, motion artefacts occur

during daily routines of the subject. To limit this chal-

lenge, we introduce the condition that measurements

are only valid if the person sits or lies. If a person

walks around, the uncertainty becomes too high, so

that no measurements are taken.

The selection of the lens is crucial at this point.

Whereas a narrow lens has less distortions and a high

spatial resolution, it is necessary to place multiple ca-

meras within one room for observation. In contrast

to that, omnidirectional cameras are equipped with

a ﬁsheye lens, which enables to monitor the whole

room with one camera. But ﬁsheye lenses bear the

problem of radial and tangential distortions. Further-

more, objects and persons are rotated in dependency

Contact-less Vital Parameter Determination: An e-Health Solution for Elderly Care

909

Figure 1: Different scenarios to determine the heart rate in home environments from left to right: Proband sits on a chair

recorded by narrow lens camera, proband sits on a chair recorded by omnidirectional camera, proband lies in a bed recorded

by monochromous camera.

of their position in the room, as can been seen in Fi-

gure 1. In order to overcome these issues and to still

use omnidirectional cameras, we suggest to use a vir-

tual perspective camera such as proposed by Meinel

et al. (Meinel et al., 2014). This has the advantage

that the rest of the algorithm can remain in its original

form.

Another challenge we are facing is darkness such

as in the night. Due to the fact that humans sleep a

considerable time every day, we would loose a lot of

information about vital parameters. In order to over-

come this issue, we propose to perform the measure-

ments in the IR-A light spectrum rather than in the

visual light spectrum. To that goal, external LEDs

in the near infra-red are applied. In Figure 1, such a

monochromous image is shown while a proband lies

in the bed. Furthermore, in the proposed algorithm

we exclude the parts of skin colour determination, the

ICA and the channel selection because we have one

instead of three channels.

2.4 Results and Discussion

In order to evaluate whether the heart rate can be de-

termined in home environments, different measure-

ments were carried out.

Due the fact that the camera position strongly in-

ﬂuences the viewing angle and the lighting conditi-

ons, different scenarios were taken into consideration,

i. e. sitting on a chair or armchair and lying in a bed.

The distance to the subjects varies from 1.5 m up to

4 m. In total, six probands were recruited for these

three scenarios. All videos were recorded in a test ﬂat

at Chemnitz University of Technology to measure un-

der realistic conditions. An Allied G201 RGB camera

with a narrow lens was mounted on the ceiling for

the measurements. For the infra-red measurements,

an Allied Prosilica GM650 monochrome camera with

external LED spotlights was used. Furthermore, the

usage of omnidirectional cameras were evaluated. A

Polar FT1 heart rate monitor was chosen as reference

system. For a quantitative accuracy analysis, the root-

mean-square error (RMSE) is chosen. The RMSE of

different videos of the same scenario can be described

by the mean RMSE.

The results are summarised in Table 1. In the sce-

nario sitting, the RMSE has a value of 1.9 BPM. The

same value is obtained for the scenario lying in the

bed. In the IR-A spectrum an RMSE of 2.6 BPM was

attained for the heart rate determination. This value

is worse in comparison to the value in the visual light

spectrum. The reason for that can be found in the

removal of the skin colour model and the ICA. Ho-

wever, this value is still accurate for the target appli-

cation ﬁeld. Moreover, it can be shown that a measu-

rement of the heart rate with omnidirectional cameras

is possible, but the value of the RMSE has increased

because of the low spatial resolution.

Table 1: RMSE in BPM for different scenarios in home en-

vironments.

Scenario RMSE

Sitting 1.9

Lying 1.9

Lying (IR-A) 2.6

Omnidirectional camera 3.1

All results clearly indicate that the proposed op-

tical heart rate determination method is suitable for

usage in the ﬁeld of home environments.

3 REMOTE RESPIRATION RATE

DETERMINATION

3.1 Related Work

A second vital parameter of interest is the respira-

tion rate. Pathological indicators of the respiration

VISAPP 2019 - 14th International Conference on Computer Vision Theory and Applications

910

rate such as breathlessness and hyperventilation can

be symptoms for several diseases. Conventional met-

hods in clinical set-ups are for example respiratory ef-

fort belts, nasal thermistors or pressure transducers.

However, all of these methods do not work contact-

less.

In this work, we want to focus on contact-less met-

hods working in the visible light spectrum. A ﬁrst

approach was presented by Tan et al., which is based

on edge detection and frame differencing (Tan et al.,

2010). Other approaches make use of autoregressive

models (Tarassenko et al., 2014) or use Eulerian video

magniﬁcation (Sharma et al., 2015).

The largest group of methods is based on the prin-

ciples of optical ﬂow. On the basis of works of Na-

kajima et al. (Nakajima et al., 2001) and Frigola et al.

(Frigola et al., 2002), Lukac et al. use a KLT tracking

to determine the optical ﬂow (Lukac et al., 2014).

3.2 Proposed Method

Based on the previous work by Wiede et al. (Wiede

et al., 2017), a new algorithm was developed in the

presented work. Firstly, images is acquired by an

RGB or a monochrome camera. For the proposed al-

gorithm it does not matter whether the images are co-

loured or grey value based, because our method relies

on the lifting and lowering of the torso that is induced

by the breathing.

In order to detect these torso motions, an ROI is

placed in the image on the chest region. To that goal,

we detect the face ﬁrstly. In comparison to Wiede

et al. (Wiede et al., 2017), we suggest to use a face

detector using the normalised pixel difference (Liao

et al., 2016), which is by far more accurate. Based

on the face region, we are able to determine the chest

ROI. Subsequently, this ROI is split in four subregions

of equal size.

In the next step, suitable features, in this case mi-

nimum Eigenvalue features (Shi and Tomasi, 1993),

are detected in the subregions. The found feature

points are tracked over time using the optical ﬂow

KLT method (Tomasi and Kanade, 1991). By obser-

ving the averaged motion of the feature points in each

of the subregions, the y-trajectory can be extracted in

a time channel. Only the y-component is considered

in the following because it has the highest contribu-

tion in the breathing motion.

Thereupon, a bandpass ﬁltering is performed to

exclude implausible frequencies. In order to reject

motion artefacts, a PCA is applied to the four ﬁltered

time channels of the subregions. In the following, the

principal component with the highest spectrum den-

sity is considered as the wanted breathing signal. The

ﬁnal respiration rate is obtained from the dominant

peak in the frequency spectrum computed with the

FFT.

3.3 Adjustments for Home

Environment Measurements

The problems that occur while determining the he-

art rate remotely in home environments (see 2.3) are

the same as for the respiration rate. Especially the

mounting of the camera on the ceiling is challenging

because it inﬂuences the detection of the face and

consequently the ROI selection. This issue can be

overcome by using a face detector of Liao et al. (Liao

et al., 2016), which can deal with large perspective

distortions.

Avoiding motion artefacts while determining the

respiration rate is more crucial than for the heart

rate because the respiration rate algorithm is motion-

based. Especially when breathing motion and arbi-

trary motion share the same frequency band, a clear

determination is not always possible. For that reason,

the inﬂuence of motion should be evaluated.

The possibilities of using omnidirectional came-

ras for respiration rate determination were investiga-

ted. For that, virtual perspective cameras, such as des-

cribed in Section 2.3, were applied as well. A deter-

mination of the respiration rate during darkness did

not require any changes in the algorithm because the

method is motion-based. To carry out the image pro-

cessing in the IR-A spectrum, only an external illumi-

nation was necessary.

3.4 Results and Discussion

The RMSE is used once more for the quantitative

assessment. For the recordings in the visible light

spectrum, a Basler acA640-100gc industrial camera

and an Allied Prosilica GM650 monochrome camera

in the IR-A spectrum were used. In total, four diffe-

rent scenarios with six probands were recorded in the

test ﬂat of Chemnitz University of Technology. As

described in Section 2.4, we evaluate the scenarios

sitting on chair or armchair, lying in the bed, during

darkness and with an omnidirectional camera.

In order to have a reference for these estimated va-

lues, we developed an own reference system, which

uses a piezo-electric transducer (MLT1132) to gather

information about the chest movement. The data eva-

luation is processed on an STM32F401RE embedded

board. Both the reference measurement and the video

recording are carried out simultaneously.

The results for the home environment measure-

ments are outlined in Table 2. It can been seen that

Contact-less Vital Parameter Determination: An e-Health Solution for Elderly Care

911

the RMSE for sitting and lying is relatively small with

1.8 breaths per minute and 2.0 breaths per minute re-

spectively. The differences might be explained by the

different viewing angles. The RMSE for the IR-A

spectrum is even lower than that. That could be due to

the fact that the proposed method is motion-based and

the probands performed less motion in darkness than

during normal lighting conditions. We expected pro-

blems because of the thick blanket over the torso for

the scenario lying in a bed, but this could not be con-

ﬁrmed. The usage of omnidirectional cameras was

also successful for determining the respiration rate re-

motely. We could show that a virtual perspective ca-

mera can compensate the disadvantages of an omni-

rectional camera.

Table 2: RMSE in breaths per minute for different scenarios

in home environments.

Scenario RMSE

Sitting 1.8

Lying 2.0

Lying (IR-A) 1.6

Omnidirectional camera 2.8

Under inclusion of all facts, there is evidence that

the respiration rate can be determined accurately in

home environments.

4 REMOTE OXYGEN

SATURATION

DETERMINATION

4.1 Related Work

The oxygen saturation in a clinical setting can be

obtained by a pulsoximeter. This measurement de-

vice determines the absorption of light emitted by

two LEDs of different wavelength and transmitted

through thin body parts such as ﬁngers or earlobes.

By calculating the so-called ratio-of-ratios at the two

wavelengths 660 nm and 940 nm for the oxygenated

and deoxygenated haemoglobin, the oxygen satura-

tion can be determined.

The basic concept of using one RGB camera to de-

termine the oxygen saturation remotely was presented

by Wieringa et al. (Wieringa et al., 2005). Tarassenko

et al. used this preliminary work to determine the oxy-

gen saturation by taking the red and blue channel of

the camera instead of taking only a small bandwidth

(Tarassenko et al., 2014). Subsequently, the AC- and

DC-parts were determined. Guazzi et al. adapted this

approach and only considered regions in the face that

have a high signal-to-noise ratio (Guazzi et al., 2015).

Alternatively, the oxygen saturation can be determi-

ned by a modulation of the reﬂected light (de Haan

and Rocque, 2015).

In contrast to that, there exists the possibility to

use two cameras in combination with optical band-

pass ﬁlters, such as proposed by Kong et al. (Kong

et al., 2013). This approach has the advantage of a

small bandwidth, but needs to match in both of the

images to analyse the same region. Verkruysse et al.

suggested to choose rather one passband of the band-

pass ﬁlters in the infra-red spectrum than in the visible

light (Verkruysse et al., 2017).

Another method is considered as an active appro-

ach because it uses LEDs of different wavelengths to

compute the oxygen saturation (Tsai et al., 2014).

Heretofore, all of the presented methods only ope-

rate under controlled environments in lab settings

with ﬁxed probands that are not moving and an active

lighting.

4.2 Proposed Method

In a ﬁrst setup, only one camera was used. Based

on the red and the blue channel of a tracked ROI, we

determined the AC- and DC-parts. Subsequently, the

ratio-of-ratio and the oxygen saturation were calcula-

ted. This approach has the disadvantage that the co-

lour channels have a small bandwidth, which leads to

inaccurate results. For that reason, we abandoned the

single camera principle.

Consequently, we developed a concept based on

two cameras. Two monochrome cameras are placed

in a stereo setup with a base distance of 4 cm with

optical bandpass ﬁlters. The wavelengths of the op-

tical bandpass ﬁlters are 660 nm and 940 nm with a

full width at half maximum of 50 nm. This is a trade-

off between a small bandwidth with a poor brightness

and a large bandwidth, which leads to an inaccurate

calculation of the ratio-of-ratios. The images of both

cameras are recorded simultaneously by means of a

trigger cable.

In our measurements, we had the goal to deter-

mine the oxygen saturation on the arm. To that goal,

at ﬁrst, a foreground segmentation was carried out to

separate foreground and background. An ROI is pla-

ced in the foreground, i. e. the arm. Subsequently,

minimum Eigenvalue features (Shi and Tomasi, 1993)

are detected and tracked with a KLT-tracker (Tomasi

and Kanade, 1991) in the video sequence. In the next

step, for each corresponding image pair, a feature ma-

tching is carried out. This guarantees that the same

ROI is considered in both images. For both ROIs one

time channel can be extracted by averaging the pixels

VISAPP 2019 - 14th International Conference on Computer Vision Theory and Applications

912

within the ROIs.

By shifting a sliding window over the time chan-

nels, the AC- and DC-parts can be computed. Hereby,

the DC value is equivalent to the mean value within

the sliding window. The gap between minimum and

maximum value in the windows represents the AC va-

lue. The ratio-of-ratio is calculated as follows:

R(φ) =

660nm

(φ)

660nm

(φ)

940nm

(φ)

940nm

(φ)

(1)

Thereby, R denotes the ratio-of-ratios and φ the cur-

rent window. Thereupon, the oxygen saturation SpO

can be determined.

SpO

= A − B · R (2)

Hereby, A and B stand for the calibration parameters,

which have to be determined during calibration.

4.3 Results and Discussion

In order to evaluate the performance of the proposed

algorithm, videos of ﬁve different probands were re-

corded. All probands were healthy and had an oxy-

gen saturation of 98-99 %. Due to the fact, that this

small range is not sufﬁcient to perform a linear ap-

proximation between ratio-of-ratios and oxygen satu-

ration, the probands had to hold their breath, which

results in a decrease of the oxygen saturation down

to 92 % at minimum. A lower value would only be

possible for ill persons with a decreased oxygen satu-

ration.

The measurements were carried out by two Basler

A640gm monochrome cameras with disabled auto-

matic white balance and disabled automatic exposure

time control. As reference system, a Pulox Po200 pul-

soximeter was used during the measurements.

The results indicate that there is a linear depen-

dency between the ratio-of-ratios and the reference

oxygen saturation. However, a robust determination

of the calibration parameters A and B is not possible

for our sequences. There are large individual differen-

ces between the single subjects. Therefore, a usage in

home environments is impossible at the moment, so

that further work has to be carried out.

However, there is a huge potential for using a re-

mote oxygen determination system. Current systems,

such as the pulsoximeter, only provide information

about the oxygen saturation on a certain body part,

e. g. the ﬁnger. Under utilisation of an image-based

system, an oxygen saturation map on the skin is fe-

asible. With such a map, it is possible to detect lo-

cal deviations of the oxygen saturation in certain body

parts. That would be a strong indicator for abnorma-

lities and diseases.

5 REMOTE BLOOD PRESSURE

DETERMINATION

5.1 Related Work

Measuring the blood pressure remotely is the most re-

cent research topic in the ﬁeld of vital parameters and

is based on the methods of remote heart rate deter-

mination. Normally, the blood pressure is measured

by means of a blood pressure cuff. This procedure is

non-invasive but not remote.

Murakami et al. proposed to use the pulse tran-

sit time (PPT) between signals extracted from a hand

and a foot ROI (Murakami et al., 2015). They used

the green channel and ﬁltered this channel to obtain

the PTT. The blood pressure can be determined be-

cause of its linear relationship to the PTT. In a similar

way, Jeong et al. (Jeong and Finkelstein, 2016) and

Secerbegood et al. (Secerbegovic et al., 2016) calcu-

lated the heart rate but used a face and a hand region.

In comparison to that, Sugita et al. suggested to de-

termine the ROIs of the right palm, the forehead and

the left cheek manually and calculate the phase dif-

ference between the two signals by using the Hilbert

transform (Sugita et al., 2015).

Although the related works demonstrated an expe-

rimental feasibility, more work has to be carried out.

5.2 Proposed Method

A direct determination of the blood pressure by me-

ans of the magnitude of the time-varying signal is im-

possible because of the variety of inﬂuencing factors,

such as lighting, exposure time, gain factor or indivi-

dual differences of the vessels.

Hence, an indirect method is advantageous. There

is a linear relationship between the blood pressure

(BP) and the pulse wave velocity (PWV). The higher

the PWV, the higher is the blood pressure. This can

be modelled by the following equation:

BP = a · PWV + b (3)

Hereby, a and b denote individual factors, which

have to determined by calibration. The PWV can be

obtained by measuring the time t

Arm

that a pulse wave

needs to pass the distance s

Arm

between two known

ROIs.

In order to obtain the PWV, the ROIs have to be

determined on the arm. Furthermore, the distance bet-

Contact-less Vital Parameter Determination: An e-Health Solution for Elderly Care

913

ween the two ROIs has to be measured. Currently, this

is done manually but should be automatised in future.

For the detection and tracking of features, we use mi-

nimum Eigenvalue features and a KLT tracking. This

is done to reduce the inﬂuence of motion artefacts.

A time signal is extracted within a single ROI by

averaging all its pixel. By detecting identical peaks

of the two regions, the transit time can be calculated.

Hereby, the frame rate has to be high enough to detect

the peaks and to determine the time difference accu-

rately.

5.3 Results and Discussion

For the evaluation, the probands have to sit still and

should not move. A Basler acA640-100gc RGB ca-

mera is placed in 0.5 m distance to the subject’s right

arm. The frame rate is ﬁxed to 50 frames per second.

For reference measurements, a boso medicus X

blood pressure cuff is used. Due to the fact that a me-

asurement with the reference system will inﬂuence a

camera based system, a synchronous measurement is

not sensible. Therefore, the blood pressure is deter-

mined with the reference system directly before and

after the image recording. If both values do not vary

too much, the recording is considered as relevant. Ot-

herwise, it is rejected. The probands had to rest at

least 10 min before the measurement starts to guaran-

tee a stable blood pressure.

First measurements in the lab proved the linear

correlation between blood pressure and PWV. Howe-

ver, more work has to be carried out to reduce the

inﬂuence of individual differences. An utilisation in

home environments is not given at the moment.

Nevertheless, this image-based method has some

advantages in comparison to the conventional met-

hod. For patients who have to measure their blood

pressure regularly (e. g. automatic measurement

every 15 or 30 min), a contact-less method provi-

des considerably more comfort. Furthermore, from

a medical point of view it is relevant, to determine the

blood pressure on different body parts and not only on

one arm. If the PWV is smaller only in one limb, this

can be an indicator for peripheral arterial diseases.

6 CONCLUSIONS

In our work we evaluated the vital parameters he-

art rate, respiration rate, oxygen saturation and blood

pressure concerning their remote, image-based deter-

mination in home environments. The results are sum-

marised in Table 3. The heart rate and the respira-

tion rate can be determined accurately and can be de-

ployed in the ﬁeld of home environments. The met-

hods to determine the oxygen saturation and the blood

pressure have to be made more robust before evalua-

ting them in real world scenarios.

Table 3: Availability of remote vital parameter determina-

tion in home environments.

Vital parameter Availability

Heart rate Ready for usage in the ﬁeld

Respiration rate Ready for usage in the ﬁeld

Oxygen saturation Improvements necessary

Blood pressure Improvements necessary

Next to the ﬁeld of home environments, remote vi-

tal parameter determination can be beneﬁcial in other

ﬁelds as well. Possible application ﬁelds are the pre-

vention of sudden infant death syndrome, sleep mo-

nitoring, monitoring of a driver’s well being, monito-

ring of training in rehabilitation centres and triage in

hospitals.

REFERENCES

Balakrishnan, G., Durand, F., and Guttag, J. (2013). De-

tecting Pulse from Head Motions in Video. In Com-

puter Vision and Pattern Recognition (CVPR), 2013

IEEE Conference on, pages 3430–3437.

Berger, R., Kaltenbach, T., B

osch, L., Erharter, M., Fath, S.,

Hosseini, M., and Magunia, P. (2016). Digital and dis-

rupted: All change for healthcare. Think Act, page 16.

de Haan, G. and Rocque, M. (2015). Device, system and

method for determining the concentration of a sub-

stance in the blood of a subject. US Patent App.

14/941,934.

Frigola, M., Amat, J., and Pags, J. (2002). Vision based

respiratory monitoring system. In Proceedings of the

10th Mediterranean Conference on Control and Auto-

mationMED2002 Lisbon, Portugal.

Garud, H., Ray, A. K., Mahadevappa, M., Chatterjee, J., and

Mandal, S. (2014). A fast auto white balance scheme

for digital pathology. In 2014 IEEE-EMBS Internati-

onal Conference on Biomedical and Health Informa-

tics, BHI 2014, pages 153–156.

Guazzi, A. R., Villarroel, M., Jorge, J., Daly, J., Frise,

M. C., Robbins, P. A., and Tarassenko, L. (2015).

Non-contact measurement of oxygen saturation with

an rgb camera. Biomedical optics express, 6(9):3320–

3338.

Hertzman, A. B. and Spealman, C. R. (1937). Observations

on the ﬁnger volume pulse recorded photoelectrically.

American Journal of Physiology, 119:334–335.

Humphreys, K., Markham, C., and Ward, T. (2005). A

CMOS camera-based system for clinical photoplet-

hysmographic applications. In Proceedings of SPIE,

volume 5823, pages 88–95.

VISAPP 2019 - 14th International Conference on Computer Vision Theory and Applications

914

Jeong, I. C. and Finkelstein, J. (2016). Introducing con-

tactless blood pressure assessment using a high speed

video camera. Journal of Medical Systems, 40(4):77.

Kong, L., Zhao, Y., Dong, L., Jian, Y., Jin, X., Li, B., Feng,

Y., Liu, M., Liu, X., and Wu, H. (2013). Non-contact

detection of oxygen saturation based on visible light

imaging device using ambient light. Optics express,

21(15):17464–17471.

Liao, S., Jain, A. K., and Li, S. Z. (2016). A Fast and Accu-

rate Unconstrained Face Detector. IEEE Transacti-

ons on Pattern Analysis and Machine Intelligence,

38(2):211–223.

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.

Meinel, L., Wiede, C., Findeisen, M., Apitzsch, A., and

Hirtz, G. (2014). Virtual perspective views for real-

time people detection using an omnidirectional ca-

mera. In Imaging Systems and Techniques (IST), 2014

IEEE International Conference on, pages 312–315.

IEEE.

Murakami, K., Yoshioka, M., and Ozawa, J. (2015). Non-

contact pulse transit time measurement using imaging

camera, and its relation to blood pressure. In 2015

14th IAPR International Conference on Machine Vi-

sion Applications (MVA), pages 414–417.

Nakajima, K., Matsumoto, Y., and Tamura, T. (2001). Deve-

lopment of real-time image sequence analysis for eva-

luating posture change and respiratory rate of a subject

in bed. Physiological Measurement, 22:N21.

Poh, M.-Z., McDuff, D., and Picard, R. (2010). Non-

contact, automated cardiac pulse measurements using

video imaging and blind source separation. Optics Ex-

press, 18(10):10762–10774.

Secerbegovic, A., Bergsland, J., Halvorsen, P., Suljanovic,

N., Mujcic, A., and Balasingham, I. (2016). Blood

pressure estimation using video plethysmography. In

2016 IEEE 13th International Symposium on Biome-

dical Imagin), pages 461–464.

Sharma, S., Bhattacharyya, S., Mukherjee, J., Purkait, P. K.,

Biswas, A., and Deb, A. K. (2015). Automated de-

tection of newborn sleep apnea using video moni-

toring system. In Advances in Pattern Recognition

(ICAPR), 2015 Eighth International Conference on,

pages 1–6. IEEE.

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

Technical report, Cornell University, Ithaca, NY,

USA.

Sugita, N., Obara, K., Yoshizawa, M., Abe, M., Tanaka, A.,

and Homma, N. (2015). Techniques for estimating

blood pressure variation using video images. In 2015

37th Annual International Conference of the IEEE En-

gineering in Medicine and Biology Society (EMBC),

pages 4218–4221.

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

Real-time vision based respiration monitoring system.

In Communication Systems Networks and Digital Sig-

nal Processing (CSNDSP), 2010 7th International

Symposium on, pages 770–774. IEEE.

Tarassenko, L., Villarroel, M., Guazzi, A., Jorge, J., Clif-

ton, 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.

Tsai, H.-Y., Huang, K.-C., Chang, H.-C., Yeh, J.-L. A.,

and Chang, C.-H. (2014). A noncontact skin oxygen-

saturation imaging system for measuring human tissue

oxygen saturation. IEEE Transactions on Instrumen-

tation and Measurement, 63(11):2620–2631.

van Gastel, M., Zinger, S., Kemps, H., and de With, P.

(2014). e-health video system for performance ana-

lysis in heart revalidation cycling. In Consumer Elec-

tronics Berlin (ICCE-Berlin), 2014 IEEE Fourth In-

ternational Conference on, pages 31–35.

Verkruysse, W., Bartula, M., Bresch, E., Rocque, M., Mef-

tah, M., and Kirenko, I. (2017). Calibration of con-

tactless pulse oximetry. Anesthesia and analgesia,

124(1):136–145.

Verkruysse, W., Svaasand, L. O., and Nelson, J. S. (2008).

Remote plethysmographic imaging using ambient

light. Optics Express, 16(26):21434–21445.

Wiede, C., Richter, J., Apitzsch, A., KhairAldin, F., and

Hirtz, G. (2016). Remote Heart Rate Determination

in RGB Data. In Proceedings of the 5th International

Conference on Pattern Recognition Applications and

Methods, pages 240–246.

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 Proceedings of

the 12th International Joint Conference on Compu-

ter Vision, Imaging and Computer Graphics Theory

and Applications - Volume 4: VISAPP, (VISIGRAPP

2017), pages 326–333. INSTICC, SciTePress.

Wiede, C., Sun, J., Richter, J., and Hirtz, G. (2018). Robust

remote heart rate determination for e-rehabilitation -

a method that overcomes motion and intensity arte-

facts. In Proceedings of the 13th International Joint

Conference on Computer Vision, Imaging and Com-

puter Graphics Theory and Applications - Volume 4:

VISAPP,, pages 491–500. INSTICC, SciTePress.

Wieringa, F. P., Mastik, F., and van der Steen, A. F.

(2005). Contactless multiple wavelength photoplet-

hysmographic imaging: a ﬁrst step toward spo2 ca-

mera technology. Annals of biomedical engineering,

33(8):1034–1041.

Zhu, X. and Ramanan, D. (2012). Face detection, pose esti-

mation, and landmark localization in the wild. In Pro-

ceedings of the IEEE Computer Society Conference

on Computer Vision and Pattern Recognition, pages

2879–2886.

Contact-less Vital Parameter Determination: An e-Health Solution for Elderly Care

915