Towards Human Posture Detection Based on Differential Measurements

Using Wearable Barometric Pressure Sensors

Nico Graum

uller

1 a

, Constantin Gis

, Franziska Geiger

2 b

, Iman Soodmand

2 c

Maeruan Kebbach

2 d

, Rainer Bader

, Christian Haubelt

1 e

and Florian Gr

utzmacher

1 f

Institute of Applied Microelectronics and Computer Engineering, University of Rostock, 18059 Rostock, Germany

Department of Orthopaedics, Rostock University Medical Center, 18057 Rostock, Germany

Keywords:

Posture Detection, Barometric Pressure Sensors, Wearable Sensors, Differential Height Measurements.

Abstract:

The detection of human postures is a well-studied research area that is closely related to human activity recog-

nition. Recent advantages of MEMS-based barometric pressure sensors have made them an interesting addi-

tional sensing modality apart from IMU-based approaches. State-of-the-art barometric pressure sensors allow

for measuring changes in barometric pressure corresponding to height differences in the range of centimeters.

However, they are susceptible to environmental pressure changes, which can signiﬁcantly inﬂuence the appli-

cation. Therefore, we propose a posture detection approach based on differential height measurements from

multiple body-worn barometric pressure sensors. We conducted an initial laboratory study with 13 subjects

(eight males and four females), evaluating standing, sitting, and lying down postures using four body-worn

barometric pressure sensors positioned at the head, hip, wrist, and ankle. Our results demonstrate that only

two sensors are needed to separate the studied postures in the feature space. Furthermore, the differential

height measurement approach can compensate for environmental pressure inﬂuences to an insigniﬁcant level

w.r.t. posture separability in our setup. The efﬁcacy of our proposed approach is further substantiated by the

observed separability of sitting on a bed and a chair for each subject individually.

1 INTRODUCTION

Barometric pressure sensors have found their way into

Human Activity Recognition (HAR) systems since

the 1990s (Manivannan et al., 2020). Due to their

ability to capture atmospheric pressure, which de-

creases with increasing altitude, they deliver valu-

able information on the altitude of a person when at-

tached or worn, e.g., in a smartwatch (Afram et al.,

2022). State-of-the-art Micro-Electro-Mechanical

Systems (MEMS) barometric pressure sensors, such

as the BMP581 from Bosch Sensortec (Bosch Sen-

sortec GmbH, 2024), can measure pressure changes

corresponding to height differences of a few cen-

timeters and are nowadays ubiquitous in devices like

smartphones or smartwatches. This makes them a

https://orcid.org/0009-0001-8914-3280

https://orcid.org/0009-0005-2473-6957

https://orcid.org/0000-0003-4919-5446

https://orcid.org/0000-0001-9564-7963

https://orcid.org/0000-0002-1568-5423

https://orcid.org/0000-0003-0370-222X

valuable option for enhancing HAR and smart health

monitoring systems in terms of detection accuracy. In

contrast to standard inertial measurement unit (IMU)-

based approaches, which are only able to detect rel-

ative movements or absolute orientations, barometric

pressure sensors are able to provide absolute height

information along the gravitational axis.

The present paper proposes human posture detec-

tion based on height differences between body parts

obtained from multiple body-worn barometric pres-

sure sensor measurements. This can be seen as a

ﬁrst step towards the aforementioned improvement of

HAR systems, which we suspect will be especially

useful for distinguishing activities that share a similar

movement proﬁle within IMU data but are performed

with different body postures.

Our paper is structured as follows: Section 2 ex-

plains related literature and how our contribution dif-

fers. Section 3 describes our proposed concept. Our

feasibility study is described in section 4 followed by

the evaluation of the acquired results in section 5. Fi-

nally, we discuss our ﬁndings in section 6 and present

conclusions in section 7.

1002

Graumüller, N., Gis, C., Geiger, F., Soodmand, I., Kebbach, M., Bader, R., Haubelt, C. and Grützmacher, F.

Towards Human Posture Detection Based on Differential Measurements Using Wearable Barometric Pressure Sensors.

DOI: 10.5220/0013273000003911

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2025) - Volume 1, pages 1002-1011

ISBN: 978-989-758-731-3; ISSN: 2184-4305

2 RELATED WORK

The related literature to our approach can be catego-

rized into three topics: general human posture detec-

tion approaches, IMU-based HAR and health moni-

toring approaches including barometric pressure sen-

sors, and applications based only on barometric pres-

sure sensors. In the following, the related literature

will be discussed in that order.

The predominant posture detection approach in

the literature is based on camera recordings. Many

different posture detection applications exist in this

ﬁeld, such as human fall detection (Sun and Wang,

2020) or posture estimation for sports activities

(Nadeem et al., 2021). For a detailed overview, read-

ers are referred to the survey (Ma et al., 2022).

However, camera-based systems are usually more

expensive and require complex camera alignment and

optical markers, which can be hidden during pos-

tures or behind objects. This limits camera-based ap-

proaches to laboratory conditions making wearable

sensor-based posture detection an interesting alterna-

tive, which is practically not limited to stationary se-

tups.

While some existing approaches incorporate only

accelerometers to detect the tiltation of the human

body for posture estimation (Chopra et al., 2016) or

to infer the posture by multiple accelerometers (Wang

et al., 2016b), others combine them with additional

sensor modalities, e.g., electromyography sensors at-

tached to the upper limb in combination with ac-

celerometers to infer the upper body posture (Li et al.,

2021). Similar to our proposed approach, some other

studies suggest to infer body postures by distance

measures between wearable sensors. In (Vecchio

et al., 2017), ultra-wideband trackers have been used

to calculate distances, and in (Matsumoto and Takano,

2016), the Received Signal Strength Indicator (RSSI)

of Bluetooth Low Energy (BLE) beacons have been

used. The survey on wearable sensor-based human

motion and posture monitoring in (Huang et al., 2023)

shows that barometric pressure sensors play no signif-

icant role in existing posture detection approaches.

However, barometric pressure sensors have al-

ready been widely investigated for tracking human ac-

tivities. The work of Manivannan et al. (Manivannan

et al., 2020) provides a good overview of the research

topics and challenges regarding the usage of baromet-

ric pressure sensors for HAR. Many approaches pre-

sented in the literature fuse sensor data of barometric

sensors with other sensing modalities like IMUs.

Most common approaches make use of additional

barometric pressure sensors in order to detect if a per-

son is inside or outside of a building and its transitions

(Zhu et al., 2020) or different modes of vertical trans-

portation, e.g., riding an elevator (Liu et al., 2018),

climbing up and down (Nam and Park, 2013; Xu and

Qiu, 2021; Leuenberger et al., 2014), or ﬂoor local-

ization (Xu et al., 2017; Liu et al., 2014). This also

includes studies on vertical velocity and height infor-

mation (Sabatini and Genovese, 2014). The same idea

was also applied to activities that involve the vertical

displacements of the entire body, i.e., falling detection

(Bianchi et al., 2010; Wang et al., 2016a; Ejupi et al.,

2017b; Pierleoni et al., 2016).

In (Ejupi et al., 2017a), (Mass

e et al., 2016), (Xie

et al., 2018), and (Mass

e et al., 2015), a single baro-

metric pressure sensor in combination with IMU sen-

sors was used to classify sit-to-stand and stand-to-

sit transitions. While this can be categorized as a

posture change detection, the single sensor approach

lacks a reference and might lead to misclassiﬁcations,

especially when sitting or lying surfaces are at dif-

ferent heights or environmental disturbances through,

for example, opening or closing doors or windows

are present. In contrast, our approach uses multiple

barometric sensors for differential pressure evalua-

tion. This leads to the detection of relative height dif-

ferences, which helps to distinguish postures indepen-

dent of the altitude at which the postures are adopted

and allows the removal of disturbances common to all

pressure sensors. This compensation approach has al-

ready been suggested in (Audisio, 2024) for tracking

the body’s center of mass during human activities, al-

though it has not been implemented and tested yet.

Other studies that involve multiple barometric

pressure sensors for HAR classiﬁcation are (Makma

et al., 2021) and (Moncada-Torres et al., 2014). While

in (Makma et al., 2021), the difference of a single

body-worn barometric pressure sensor to a stationary

wall-mounted reference sensor is utilized, (Moncada-

Torres et al., 2014) used the altitude changes relative

to the individual arithmetic mean of multiple body-

worn barometric pressure sensors. In contrast, our

approach utilizes the relative differences between all

body-worn barometric pressure sensors to cancel out

environmental inﬂuences affecting all sensors and in-

fer the adopted body posture.

Finally, examples of applications that use baro-

metric pressure sensors only include altimetry mea-

surement systems (Bolanakis, 2017b; Bolanakis,

2017a) and the detection of opening and clos-

ing events of doors inside a building (Wu et al.,

2015). Furthermore, in (Bollmeyer et al., 2013) and

(Bollmeyer et al., 2014), altitude information of a per-

son in medical applications is calculated based on dif-

ferential measurements. Again, only one body-worn

sensor is used together with a stationary reference

Towards Human Posture Detection Based on Differential Measurements Using Wearable Barometric Pressure Sensors

1003

measurement. In (Vanini et al., 2016), the barometric

pressure sensor of a smartphone is used to distinguish

between standing/walking on level ground and climb-

ing stairs, riding an elevator, and riding a cable car.

Ghimire et al. (Ghimire et al., 2016) studied the clas-

siﬁcation of motion states (resting, walking, riding an

elevator) using a single body-worn barometric pres-

sure sensor. Similarly, in (Mass

e et al., 2014), a single

trunk-worn barometric pressure sensor is used to dis-

tinguish standing and sitting activities. The most sim-

ilar approach regarding the measurement of differen-

tial pressure between multiple body-worn barometric

pressure sensors was proposed by Sun et al. in (Sun

et al., 2019). They used two body-worn barometric

pressure sensors to detect a person’s falling in differ-

ent scenarios by calculating the height information of

a waist-worn sensor to a reference sensor attached to

the shoes. Although they state that the reference sen-

sor can be mounted stationary or removed completely,

the detection accuracy may drop. This substantiates

the potential of our proposed approach to differential

barometric pressure measurements for detecting hu-

man postures, which can be seen as an extension of

the aforementioned approaches towards detecting hu-

man postures.

3 CONCEPT

As described earlier, our approach is based on multi-

ple barometric pressure sensors attached to different

body parts, which gives us the following advantages:

Differential Height Measurements: between the

sensors can be extracted from the minor relative

changes in barometric pressure due to different pos-

tures. Recent MEMS-based sensors enable us to mea-

sure pressure changes corresponding to a few cen-

timeters, which lies in the range needed to detect

height differences between body parts.

Compensation of Environmental Inﬂuences: will

be enabled due to the correlation of disturbances in all

accelerometers. Besides useful information regard-

ing the pressure on different body parts, the data also

contains disturbances induced by events in the envi-

ronment of the measurement setup. These could be

caused by opening or closing doors or windows in

the building or by barometric pressure changes due

to weather conditions like wind. We suspect that

these events inﬂuence the barometric pressure mea-

surements of all sensors similarly, enabling the elimi-

nation by considering only the height differences.

(a)

(b)

Figure 1: (a) Examined body postures standing, sitting, and

lying (b) Visualization of the placement of the four body-

worn sensors (red) at the head, wrist, hip, and ankle as well

as the reference sensor placed on the ground (blue). The

reference sensor is not involved in the posture detection but

is used to evaluate environmental inﬂuences in barometric

pressure. (Created in BioRender. Geiger, F. (2024) https:

//BioRender.com/f38h800).

Table 1: Execution sequence of the user study.

1. Standing 6. Sitting on a bed

2. Sitting on a chair 7. Lying down on a bed

3. Standing 8. Sitting on a bed

4. Lying down on a bed 9. Standing

5. Standing

4 EXPERIMENTS

4.1 User Study

The aim of the experiment is to assess how well a

person’s posture can be inferred from multiple body-

worn barometric pressure sensors. To this end, our

study was conducted under laboratory conditions,

minimizing additional environmental inﬂuences as

much as possible. Therefore, the experiments were

conducted in a computer lab on the second ﬂoor of a

4-ﬂoor building with an approximate area of 85 m

and a ceiling height of 3.30 m. The doors and win-

dows were kept closed during the experiments, and

the air conditioning was turned off to minimize envi-

ronmental inﬂuences on the barometric pressure. The

experiments were conducted from 10 am to 4 pm,

and the room temperature varied between 21.1

◦

C and

21.5

◦

C during the experiment.

The study involved 13 healthy subjects (eight

males and ﬁve females) with an average age of

BIOSIGNALS 2025 - 18th International Conference on Bio-inspired Systems and Signal Processing

1004

34.6±9.6 years. Ethical approval was granted by the

Ethics Committee of the Rostock University Medi-

cal Center, Germany (A 2024-0138). The average

body height was 181±13 cm, with the tallest subject

measuring 208 cm and the shortest subject measuring

161 cm. The subjects were instructed to follow a pre-

deﬁned sequence of nine postures speciﬁed in Table

1, which involved standing, sitting on a chair, sitting

on a bed, and lying on a bed. In our setup, a ﬁeld cot

served as the bed. The postures are displayed in Fig-

ure 1a. In order to capture execution variations of the

same subject, the sequence was repeated three times

per subject, and each posture was held for approxi-

mately 10 s. In the ﬁrst trial, no speciﬁc instructions

were given on how to sit, stand, or lay down. For the

second and the third trial, the following instructions

were given to the subject:

• standing straight with hands hanging down beside

the body

• sitting straight with hands on the knees

• lying ﬂat on the bed with the hands resting beside

the body

The chair and the ﬁeld cot were placed close to each

other to minimize transition times. Measured from

the ﬂoor, the height of the chair was 48 cm, and the

height of the ﬁeld cot was approximately 40 cm.

4.2 Hardware Setup

The hardware setup for our study consisted of four

wearable sensor nodes attached to the human body

and a single sensor node on a ﬁxed location on the

ﬂoor within a perimeter of 2 m from the subject. The

latter was implemented to analyze environmental in-

ﬂuences on the barometric pressure data unaffected

by the subject’s movement. Note that the pressure

data from the reference sensor is not required or used

for our posture detection, which permits mobile ap-

plications. As can be seen in Figure 1b, the sensors

were attached to the head, hip, wrist, and ankle using

hook-and-loop fasteners. Each sensor node consists

of a BMP581 barometric pressure sensor from Bosch

Sensortec GmbH, Reutlingen, Germany. The sensor

nodes attached to the hip and the head, as well as the

reference sensor on the ﬂoor (in the following referred

to as hub nodes), were additionally equipped with

a microcontroller board based on an nRF52832 mi-

crocontroller with integrated BLE chip from Nordic

Semiconductors. The sensor nodes at the ankle and

wrist were wired to the hub nodes at the hip and the

head, respectively. The hub nodes sampled the baro-

metric pressure sensors via I²C at a rate of 12 Hz and

sent the data wirelessly to a laptop via BLE. Each

barometric pressure sample was transmitted in an in-

dividual BLE packet. The Unix timestamp on the lap-

top at the arrival of each BLE packet was used as the

timestamp for each pressure sample.

To annotate the data with corresponding ground

truth labels, we marked the recorded postures by

implementing two recording states, switched via a

button press. The state (posture or transition) was

recorded together with the barometric pressure data

and their timestamps and was used to distinguish

between posture and transitions. Together with the

known execution sequence of postures, all posture

states were annotated with their corresponding label

in post-processing. It is important to note that the in-

structor ﬁrst switched the recording state from posture

to transition before instructing the subject to change

to the next posture in the sequence and switched back

to the posture recording state after the subject took its

instructed position. This way, no transitions are con-

tained in the posture-labeled data.

4.3 Data Processing

Due to manufacturing tolerances, there are variations

in barometric pressure measurements between differ-

ent sensors, even when placed at the exact altitudes.

To compensate for these variations, we calibrated the

sensor nodes before each recording of a subject. Af-

terwards, the data was resampled by linear interpola-

tion, and the height differences were calculated.

4.3.1 Calibration

Before each subject’s trial, all sensor nodes were

placed on the ground, and a short sequence of pres-

sure values was collected. The mean pressure val-

ues over the recording were calculated for each sen-

sor, and the absolute differences to the ankle sensor

were calculated. This absolute difference was then

subtracted from the measured pressure sequences in

all other sensors to compensate for absolute pressure

offsets. This calibration between the sensors attached

to the body allows for posture detection independent

of any stationary sensors.

4.3.2 Height Differences

In order to calculate height differences between the

sensor nodes, ﬁrst their absolute altitude w.r.t. sea

level has been calculated by the barometric formula

using equation (1), with subscript b denoting values

for the lowest atmospheric layer from 0-11,000 m

above sea level, P

being the static pressure at sea

level (here: P

= 1, 013.25 hPA), T

being the stan-

dard temperature at sea level (here: T

= 288.15 K),

Towards Human Posture Detection Based on Differential Measurements Using Wearable Barometric Pressure Sensors

1005

being the standard temperature lapse rate (here:

= 0.0065

), h

being the height at the bottom of

atmospheric layer (here: h

= 0 m), R the universal

gas constant (R = 8.3145

mol·K

), g

gravitational ac-

celeration (g

= 9.80665

), and M molar mass of air

(M = 0.0289644

mol

h = h





1 −





−R·L

·M





(1)

From the absolute altitude values, height differences

between the individual sensor nodes have been calcu-

lated.

Since no additional synchronization between sen-

sor hub nodes was implemented, barometric pres-

sure data was sampled equidistantly but not necessar-

ily at the exact same time between different sensor

hub nodes. To calculate height differences at speciﬁc

timestamps, the absolute height measurements from

all sensors were resampled at a common time base

using linear interpolation between samples.

5 EVALUATION

The raw pressure values were calibrated to the ankle

sensor as described in section 4.3.1 and are displayed

in Figure 2 for trial 1 of subject 2. The sections high-

lighted in green mark the transition phases when the

subject changes between the postures. Each colored

line denotes one of the pressure sensor readings, and

the transparent numbers denote the execution order

according to Table 1. Note that the reference sen-

sor (purple) is only included for the visual analysis

of environmental pressure ﬂuctuations. As this sen-

sor is stationary on the ﬂoor, the measured pressure

is expected to be constant during the trial. However,

throughout the trial, the graph shows larger bumps

from approximately 5-25 s, 45-55 s, and 110-135 s.

We assume this is caused by events in the building

that were outside of our control, e.g., opening/closing

doors or windows in other rooms or corridors. The

pressure changes due to posture changes, e.g., be-

tween phases 5, 6, and 7 (standing, sitting, and lying

down), are in the same order of magnitude, further

substantiating our motivation for using height differ-

ences between sensors for the classiﬁcation. While

the observed disturbances signiﬁcantly inﬂuence the

absolute pressure, the relative pressure differences be-

tween the sensors remain nearly the same.

Figure 3 is similar to Figure 2 except that each

colored line now represents the height differences be-

tween the sensors attached to subject 2 during trial 1.

It can be seen that the calculation of height differences

compensates external inﬂuences in barometric pres-

sure to the greatest extent (cf. Figure 2). This shows

that the external variations inﬂuence all attached sen-

sors similarly. Overall, minor ﬂuctuations remain vis-

ible in all graphs, likely caused by the noise of the

sensors and insufﬁcient sensor synchronization. Nev-

ertheless, these ﬂuctuations are signiﬁcantly smaller

than the differences in height caused by different pos-

tures.

Similarly to Figure 2, the postures can be visually

inferred from the graphs. As expected, the height dif-

ferences between head and ankle, wrist and ankle, and

hip and ankle decrease when transitioning from stand-

ing (phases 1, 3, and 5) to sitting (phases 2, 6, and 8)

and to lying down (phases 4 and 7). The height dif-

ference between head and hip and between head and

wrist remains nearly the same between standing and

sitting but decreases when lying down. In contrast,

the height difference between hip and wrist does not

substantially change between the postures. Interest-

ingly, minor differences are visible between phases 2

and 6, where the subject sat on the chair and the bed,

respectively. As the bed’s surface was lower than the

chair’s surface, the height difference between hip and

ankle and between head and ankle are less for sitting

on the bed (phases 6 and 8).

To select the most suitable sensors for posture

detection, the distributions of the height differences

between all body-worn sensors are depicted as his-

tograms in Figure 4. The x-axis marks the distance

calculated for the corresponding sensor combination,

the y-axis shows the number of occurrences in our

dataset, and the color denotes the corresponding pos-

tures: standing (blue), sitting (orange and green), and

lying down (red). Looking at the distribution of the

height difference between wrist and hip, a straightfor-

ward separation of the postures is impossible, as the

distributions for the different postures signiﬁcantly

overlap. In contrast, the distributions of the height

differences between head and ankle and between hip

and ankle show a clear separation between the pos-

tures. This could allow for a simple classiﬁcation us-

ing a threshold even without normalizing the sensor

height differences to the corresponding body height.

Similar to before, when looking at the difference in

height between head and ankle and between hip and

ankle, minor differences are visible between distribu-

tions for sitting on the chair and sitting on the bed, al-

though they have a signiﬁcant overlap. The difference

between the distributions becomes even more evident

when looking at the subjects individually. This is an

interesting result as the height difference between the

chair and the bed was only 8 cm. However, we sus-

pect that discrimination in non-laboratory conditions,

BIOSIGNALS 2025 - 18th International Conference on Bio-inspired Systems and Signal Processing

1006

Figure 2: Pressure values calibrated to the ankle sensor for subject 2 in trial 1.

Figure 3: Height differences between the sensors calculated for subject 2 in trial 1.

Table 2: Values of

for each possible combination of sen-

sors.

Sensor Combination

Head and Ankle 2, 247 · 10

Head and Hip 386 · 10

Head and Wrist 478 · 10

Wrist and Hip 20 · 10

Wrist and Ankle 487 · 10

Hip and Ankle 495 · 10

where the subjects might sit differently (e.g., straight

vs. laid back), will be more difﬁcult.

To determine a quantitative measure of posture

distinguishability, we utilized a metric that quantiﬁes

the separability of data distributions, which is adopted

from multiclass linear discriminant analysis (Tharwat

et al., 2017). To this end, each posture (standing,

sitting, lying down) is considered a class. The vari-

ance between the classes S

and the variance within

the classes S

are calculated, and their ratio, which is

subject to maximization in linear discriminant analy-

sis, is used as a measure of separability. We calculated

and S

for the one-dimensional height differences

with the following equations, where X is the set of

samples x ∈ X, C is the set of classes c ∈ C, n

is the

number of samples belonging to c, N is the total num-

ber of samples N =

∑

c∈C

, µ

is the class mean of

all samples belonging to c with

∑

j∈c

and µ is the mean of all samples

µ =

∑

x∈X

x .

∑

c∈C

(µ

− µ)

(2)

∑

c∈C

∑

j∈c

− µ

)

(3)

We used

as a measure of separability between

the posture classes for each sensor combination. Note

that the measured height differences per subject were

normalized for the calculation by the overall height of

the corresponding subject to compensate for different

subject heights. The calculated values for all sensor

combinations are shown in Table 2.

Towards Human Posture Detection Based on Differential Measurements Using Wearable Barometric Pressure Sensors

1007

Figure 4: Distribution of postures over all height differences between the sensors.

As expected, the highest value is obtained for the

height difference between the head and the ankle. The

second highest value is obtained for the height dif-

ference between hip and ankle, although the height

difference between wrist and ankle and between head

and wrist have similar values. The lowest value was

obtained for the height difference between the wrist

and the hip, which aligns with our initial visual anal-

ysis. This supports our previous assumption that the

height differences between head and ankle and be-

tween hip and ankle are best suited for separating the

postures.

Figure 5 displays the distributions for each pos-

ture along the height difference between head and

ankle (x-axis) and between hip and ankle (y-axis)

in a two-dimensional feature space. Both differ-

ences were scaled per subject according to their body

height, which causes slightly more compact distri-

butions. Consistently with the previous results, the

posture classes standing (blue), sitting (orange and

green), and lying (red) are clearly separated in the

feature space, which suggests a simple classiﬁcation

using predeﬁned thresholds or simple algorithms like

decision trees or a k-nearest-neighbor. Looking at

the separability of sitting on a chair and sitting on a

bed, we can see differences in the centroids of the

two-dimensional distributions. Nevertheless, there is

a signiﬁcant overlap between the two distributions,

which would lead to a certain amount of misclassiﬁ-

cations. Similar to previous analyses, the examination

of the graphs for individual subjects indicates better

separation of the distributions between sitting on the

chair and the bed. Although we have scaled the height

differences with the body height, the overlaps could

be caused by variations due to the sensor placement

on the subjects and the movement of the sensor dur-

ing the experiments.

BIOSIGNALS 2025 - 18th International Conference on Bio-inspired Systems and Signal Processing

1008

Figure 5: Distribution of the postures over the height differ-

ences between head and ankle as well as hip and ankle.

6 DISCUSSION

Our work presents a feasibility study targeting human

posture detection using barometric pressure sensors.

The novelty lies in using the differential altitudes be-

tween multiple barometric pressure sensors, which

are all directly attached to the subject. However, our

feasibility study has several limitations which require

further investigation.

Firstly, only three distinct static postures were

evaluated. Deciding between more similar postures

will be more challenging, but our analysis showed dif-

ferences between sitting on the bed and the chair, in-

dicating the signiﬁcant potential of our approach.

Secondly, our differential approach showed ro-

bustness against the air pressure disturbances present

in our laboratory data, consistent with earlier

studies on barometer-based indoor ﬂoor localiza-

tion (Liu et al., 2014) and indoor altitude estima-

tion (Bollmeyer et al., 2014). Nevertheless, the effects

of more severe environmental disturbances of the air

pressure, as expected in real-world scenarios, require

further investigation.

Another limitation of our study is the uncompen-

sated pressure drift of the barometric pressure sen-

sors and the simplistic time synchronization between

the sensors. While the recordings were relatively

short in our experiments, the effects mentioned above

might lead to inaccuracies during long-term moni-

toring, which needs additional compensation. More-

over, our feasibility study excluded the classiﬁcation

of transitions between postures, which would require

more sophisticated algorithms and introduce addi-

tional challenges, e.g., due to overlaps in the tran-

sitions. Also, the selected postures were limited in

their execution variation, and postures during activ-

ities, e.g., upright posture in activities like walking

or running, were not considered. This could intro-

duce additional disturbances, such as local baromet-

ric pressure changes, due to air compression in front

of a person. The inﬂuence on the local barometric

pressure is currently unknown to us. Still, we suspect

that the differential height approach might also help to

compensate for this effect under the assumption that

all sensor nodes are equally affected by this.

The abovementioned limitations highlight that our

work presents the ﬁrst results towards a robust pos-

ture detection approach using the height differences

measured with multiple body-worn barometric pres-

sure sensors. However, the observed separability be-

tween the examined postures and the observed com-

pensatory ability of differential measurements in our

experiments motivate further research in this direc-

tion. Furthermore, contrary to standard IMU sensors,

which are only able to detect relative movements or

absolute orientations, our approach is able to deliver

an absolute height difference along the gravitational

axis.

7 CONCLUSION

In the paper at hand, we propose a posture detection

approach based on differential height measurements

from wearable barometric pressure sensors. We eval-

uated our approach in an initial user study with 13

subjects under laboratory conditions, examining the

postures of standing, sitting, and lying down on a bed

with barometric pressure sensors attached to the head,

hip, wrist, and ankle of each subject. Our evaluation

showed that a combination of two sensors is sufﬁcient

to discriminate all examined postures in our experi-

ment. Furthermore, the proposed differential height

measurement approach successfully compensated for

environmental inﬂuences on the barometric pressure

sufﬁciently in our study. The separability of each sub-

ject’s sitting posture at different surface heights fur-

ther substantiates our proposed posture detection ap-

proach. In future work, we will concentrate on en-

hancing the robustness of our approach w.r.t. envi-

ronmental inﬂuences in real-life scenarios, as well as

its potential application to health monitoring appli-

cations, e.g., the estimation of joint angles. Further-

more, the improvement in distinguishability of activ-

ities in HAR systems by accurate posture detection

using barometric pressure sensors remains to be eval-

uated.

Towards Human Posture Detection Based on Differential Measurements Using Wearable Barometric Pressure Sensors

1009

ACKNOWLEDGEMENTS

This work is partially funded by the German Sci-

ence Foundation (DFG) - SFB 1270/2 - 299150580.

Furthermore, we acknowledge ﬁnancial support from

the department ”Aging of the Individual and Soci-

ety” (AGIS, German: Altern des Individuums und der

Gesellschaft) of the University of Rostock, Germany,

due to the special dividend for smart health projects.

Lastly, we thank Bosch Sensortec GmbH, Germany,

for providing the required sensor hardware.

INSTITUTIONAL REVIEW BOARD

STATEMENT

The study was conducted in accordance with the Dec-

laration of Helsinki and approved by the Ethics Com-

mittee of the Medical Faculty, University of Rostock

(registration number A 2024-0138).

REFERENCES

Afram, R. M., Abd Al-Muhsen, G. S., Ali, Y. H., Bashar,

A.-S., Ahmed, W. R., Ali, H. A., Jasim, A. A.,

Josephine, R., Muslim, A. M., and Ali, E. H. (2022).

Real-time motion up and down activity recognition

based on smart phone and smart watch sensors. In

2022 4th International Conference on Current Re-

search in Engineering and Science Applications (IC-

CRESA), pages 240–244. IEEE.

Audisio, A. (2024). Tracking the Body Center of Mass

during simulated daily activities: a sensor fusion ap-

proach with barometric and inertial data. Phd thesis,

Politecnico di Torino.

Bianchi, F., Redmond, S. J., Narayanan, M. R., Cerutti, S.,

and Lovell, N. H. (2010). Barometric pressure and tri-

axial accelerometry-based falls event detection. IEEE

Transactions on Neural Systems and Rehabilitation

Engineering, 18(6):619–627.

Bolanakis, D. E. (2017a). Evaluating performance of

MEMS barometric sensors in differential altimetry

systems. IEEE Aerospace and Electronic Systems

Magazine, 32(9):34–39.

Bolanakis, D. E. (2017b). MEMS Barometers Toward Verti-

cal Position Detection: Background Theory, System

Prototyping, and Measurement Analysis. Synthesis

Lectures on Mechanical Engineering. Springer Inter-

national Publishing, Cham.

Bollmeyer, C., Esemann, T., Gehring, H., and Hellbr

uck,

H. (2013). Precise indoor altitude estimation based

on differential barometric sensing for wireless medi-

cal applications. In 2013 IEEE International Confer-

ence on Body Sensor Networks, pages 1–6.

Bollmeyer, C., Pelka, M., Gehring, H., and Hellbr

uck, H.

(2014). Evaluation of radio based, optical and baro-

metric localization for indoor altitude estimation in

medical applications. In 2014 International Confer-

ence on Indoor Positioning and Indoor Navigation

(IPIN), pages 132–141. IEEE.

Bosch Sensortec GmbH (2024). BMP581 – Barometric

Pressure Sensor. Revision 1.9.

Chopra, S., Kumar, M., and Sood, S. (2016). Wearable pos-

ture detection and alert system. In 2016 International

Conference System Modeling & Advancement in Re-

search Trends (SMART), pages 130–134. IEEE.

Ejupi, A., Brodie, M., Lord, S. R., Annegarn, J., Red-

mond, S. J., and Delbaere, K. (2017a). Wavelet-

Based Sit-To-Stand Detection and Assessment of Fall

Risk in Older People Using a Wearable Pendant De-

vice. IEEE Transactions on Biomedical Engineering,

64(7):1602–1607.

Ejupi, A., Galang, C., Aziz, O., Park, E. J., and Robi-

novitch, S. (2017b). Accuracy of a wavelet-based

fall detection approach using an accelerometer and

a barometric pressure sensor. In 2017 39th Annual

International Conference of the IEEE Engineering in

Medicine and Biology Society (EMBC), pages 2150–

2153. IEEE.

Ghimire, B., Nickel, C., and Seitz, J. (2016). Pedestrian

motion state classiﬁcation using pressure sensors. In

2016 International Conference on Indoor Positioning

and Indoor Navigation (IPIN), pages 1–6.

Huang, X., Xue, Y., Ren, S., and Wang, F. (2023). Sensor-

based wearable systems for monitoring human motion

and posture: A review. Sensors, 23(22):9047.

Leuenberger, K., Gonzenbach, R., Wiedmer, E., Luft, A.,

and Gassert, R. (2014). Classiﬁcation of stair ascent

and descent in stroke patients. In 2014 11th Interna-

tional Conference on Wearable and Implantable Body

Sensor Networks Workshops, pages 11–16. IEEE.

Li, X., Zhou, Z., Wu, J., and Xiong, Y. (2021). Human

posture detection method based on wearable devices.

Journal of healthcare engineering, 2021(1):8879061.

Liu, K., Wang, Y., and Wang, J. (2014). Differential baro-

metric altimetry assists ﬂoor identiﬁcation in wlan lo-

cation ﬁngerprinting study. Principle and Application

Progress in Location-Based Services, pages 21–29.

Liu, M., Li, H., Wang, Y., Li, F., and Chen, X. (2018).

Double-windows-based motion recognition in multi-

ﬂoor buildings assisted by a built-in barometer. Sen-

sors, 18(4):1061.

Ma, N., Wu, Z., Cheung, Y.-m., Guo, Y., Gao, Y., Li, J., and

Jiang, B. (2022). A survey of human action recog-

nition and posture prediction. Tsinghua Science and

Technology, 27(6):973–1001.

Makma, J., Thanapatay, D., Isshiki, T., Chinrungru-

eng, J., and Thiemjarus, S. (2021). Enhanc-

ing Accelerometer-based Human Activity Recogni-

tion with Relative Barometric Pressure Signal. In

2021 18th International Conference on Electrical

Engineering/Electronics, Computer, Telecommunica-

tions and Information Technology (ECTI-CON), pages

529–532.

Manivannan, A., Chin, W. C. B., Barrat, A., and Bouffanais,

R. (2020). On the Challenges and Potential of Using

BIOSIGNALS 2025 - 18th International Conference on Bio-inspired Systems and Signal Processing

1010

Barometric Sensors to Track Human Activity. Sen-

sors, 20(23):6786.

Mass

e, F., Bourke, A. K., Chardonnens, J., Paraschiv-

Ionescu, A., and Aminian, K. (2014). Suitability of

commercial barometric pressure sensors to distinguish

sitting and standing activities for wearable monitor-

ing. Medical Engineering & Physics, 36(6):739–744.

Mass

e, F., Gonzenbach, R., Paraschiv-Ionescu, A., Luft,

A. R., and Aminian, K. (2016). Wearable Baromet-

ric Pressure Sensor to Improve Postural Transition

Recognition of Mobility-Impaired Stroke Patients.

IEEE transactions on neural systems and rehabilita-

tion engineering: a publication of the IEEE Engineer-

ing in Medicine and Biology Society, 24(11):1210–

1217.

Mass

e, F., Gonzenbach, R. R., Arami, A., Paraschiv-

Ionescu, A., Luft, A. R., and Aminian, K. (2015).

Improving activity recognition using a wearable baro-

metric pressure sensor in mobility-impaired stroke pa-

tients. Journal of NeuroEngineering and Rehabilita-

tion, 12:72.

Matsumoto, M. and Takano, K. (2016). A posture detec-

tion system using consumer wearable sensors. In 2016

10th International Conference on Complex, Intelli-

gent, and Software Intensive Systems (CISIS), pages

526–531. IEEE.

Moncada-Torres, A., Leuenberger, K., Gonzenbach, R.,

Luft, A., and Gassert, R. (2014). Activity classiﬁca-

tion based on inertial and barometric pressure sensors

at different anatomical locations. Physiological Mea-

surement, 35(7):1245.

Nadeem, A., Jalal, A., and Kim, K. (2021). Automatic

human posture estimation for sport activity recog-

nition with robust body parts detection and entropy

markov model. Multimedia Tools and Applications,

80:21465–21498.

Nam, Y. and Park, J. W. (2013). Child activity recogni-

tion based on cooperative fusion model of a triax-

ial accelerometer and a barometric pressure sensor.

IEEE journal of biomedical and health informatics,

17(2):420–426.

Pierleoni, P., Belli, A., Maurizi, L., Palma, L., Pernini,

L., Paniccia, M., and Valenti, S. (2016). A Wear-

able Fall Detector for Elderly People Based on AHRS

and Barometric Sensor. IEEE Sensors Journal,

16(17):6733–6744.

Sabatini, A. M. and Genovese, V. (2014). A Sensor Fu-

sion Method for Tracking Vertical Velocity and Height

Based on Inertial and Barometric Altimeter Measure-

ments. Sensors, 14(8):13324–13347.

Sun, G. and Wang, Z. (2020). Fall detection algorithm for

the elderly based on human posture estimation. In

2020 Asia-Paciﬁc Conference on Image Processing,

Electronics and Computers (IPEC), pages 172–176.

IEEE.

Sun, Y., Lo, F. P.-w., Thiemjarus, S., and Lo, B. (2019).

Wearable fall detection system using barometric pres-

sure sensors and machine learning. In SENSOR-

COMM 2019: The Thirteenth International Confer-

ence on Sensor Technologies and Applications.

Tharwat, A., Gaber, T., Ibrahim, A., and Hassanien, A. E.

(2017). Linear discriminant analysis: A detailed tuto-

rial. AI communications, 30(2):169–190.

Vanini, S., Faraci, F., Ferrari, A., and Giordano, S. (2016).

Using barometric pressure data to recognize vertical

displacement activities on smartphones. Computer

Communications, 87:37–48.

Vecchio, A., Mulas, F., and Cola, G. (2017). Posture recog-

nition using the interdistances between wearable de-

vices. IEEE Sensors Letters, 1(4):1–4.

Wang, C., Lu, W., Narayanan, M. R., Chang, D. C. W.,

Lord, S. R., Redmond, S. J., and Lovell, N. H.

(2016a). Low-power fall detector using triaxial ac-

celerometry and barometric pressure sensing. IEEE

Transactions on Industrial Informatics, 12(6):2302–

2311.

Wang, J., Huang, Z., Zhang, W., Patil, A., Patil, K., Zhu,

T., Shiroma, E. J., Schepps, M. A., and Harris, T. B.

(2016b). Wearable sensor based human posture recog-

nition. In 2016 IEEE International conference on big

data (big data), pages 3432–3438. IEEE.

Wu, M., Pathak, P. H., and Mohapatra, P. (2015). Moni-

toring building door events using barometer sensor in

smartphones. In Proceedings of the 2015 ACM Inter-

national Joint Conference on Pervasive and Ubiqui-

tous Computing, UbiComp ’15, pages 319–323, New

York, NY, USA. Association for Computing Machin-

ery.

Xie, L., Tian, J., Ding, G., and Zhao, Q. (2018). Human ac-

tivity recognition method based on inertial sensor and

barometer. In 2018 IEEE International Symposium on

Inertial Sensors and Systems (INERTIAL), pages 1–4.

Xu, Y. and Qiu, T. T. (2021). Human activity recogni-

tion and embedded application based on convolutional

neural network. Journal of Artiﬁcial Intelligence and

Technology, 1(1):51–60.

Xu, Z., Wei, J., Zhu, J., and Yang, W. (2017). A robust

ﬂoor localization method using inertial and barome-

ter measurements. In 2017 International Conference

on Indoor Positioning and Indoor Navigation (IPIN),

pages 1–8. IEEE.

Zhu, Y., Luo, H., Zhao, F., and Chen, R. (2020). In-

door/outdoor switching detection using multisensor

densenet and lstm. IEEE Internet of Things Journal,

8(3):1544–1556.

Towards Human Posture Detection Based on Differential Measurements Using Wearable Barometric Pressure Sensors

1011