SleepPal: A Sleep Monitoring System for Body Movement and Sleep

Posture Detection

Ali Ibrahim

1,2 a

, Kabalan Chaccour

1,2 b

, Amir Hajjam El Hassani

1 c

and Emmanuel Andres

3 d

Nanomedicine, Imagery and Therapeutic (NIT Lab.), University of Bourgogne Franche-Compt

e (UBFC), Belfort, France

TICKET Lab. Dept. of Computer and Communications Engineering, Antonine University (UA), Hadat-Baabda, Lebanon

Health Sciences Pedagogy Center, University of Strasbourg, Strasbourg, France

Keywords:

Sleep, Sleep Posture, Nocturnal Body Movements, Sleep Quality, Actigraphy.

Abstract:

Sleep posture is a clinical relevant parameter for it is associated with several pathologies and affects the

quality of sleep. In this paper, we propose SleepPal a sleep monitoring system for body movement and sleep

posture detection. It consists of a wearable device that extracts data from a 3-axis accelerometer and transmits

them to a remote monitoring station. A threshold-based algorithm is used to detect body movement and to

distinguish between transitions. The proposed system will also evaluate the sleep quality index. Experiments

were conducted on 10 subjects and results showed 88% of sensitivity and 82% of accuracy.

1 INTRODUCTION

Movements during sleep and body postures were re-

ported to be associated with different pathologies

(Horne et al., 2002) and decreased sleep quality

(De Koninck et al., 1983). For example, patients with

insomnia spend more time on their back (De Koninck

et al., 1983), whereas patients with heart failure pre-

fer to sleep on their side (Leung et al., 2003) (Hoff-

stein, 1996). Sleep apnea is also increased by certain

sleep positions such as supine position (Cartwright,

1984). Obviously, the number of pathologies that im-

plies sleep disorders has become of signiﬁcance im-

portance that it became prime to evaluate the quality

of sleep. Accurate measurement of sleep quality is

performed normally by overnight polysomnography

(PSG) which includes several physiological measure-

ments such as Electrocardiogram (ECG), Electroen-

cephalogram (EEG), Electromyogram (EMG), res-

piration and body movement during sleep (Colman,

2006). PSG is a reliable method used in sleep diagno-

sis, but it is not without drawbacks. It involves high

costs associated with the utilisation of complex equip-

ment and require continuous monitoring from health-

care professionals. In addition, attaching many elec-

https://orcid.org/0000-0003-0092-7247

https://orcid.org/0000-0002-3731-7787

https://orcid.org/0000-0002-8470-806X

https://orcid.org/0000-0002-7914-7616

trodes to the patient’s body is considered intrusive,

which can disturb sleep. Thus, the measured data

may not accurately represent the sleep behavior of

the patients. These drawbacks make PSG impractical

to be implemented within a long-term sleep monitor-

ing system within homes. Alternately, actigraphy has

been recently adopted for continuous measurement of

sleep activity. The device is called an actigraph. It is

composed of motion sensors such as accelerometers.

It is capable of measuring and logging the movement.

The advantages of actigraphy over PSG are many, to

note the cost, the low number of sensors, the mini-

mum intrusiveness and and the continuous log over

long periods of time (i.e Weeks, Months).

In this paper, a complete monitoring system in-

cluding a wearable device for a long-term sleep mon-

itoring is proposed. The wearable device named

SleepPal is comfortable to wear with low intrusive-

ness for the patient and its deployment does not re-

quire any intervention from a trained expert. It ex-

tracts data from a 3-axis accelerometer and transmits

them to a remote monitoring station over a wireless

network connection. The body posture, the position,

and the movements can be determined based on mul-

tiple features extracted from the acceleration data. A

simple but accurate threshold-based algorithm is de-

veloped that detects the transitions between different

body postures that are deﬁned in a state diagram. Fi-

nally, the sleep quality indicator will be determined

Ibrahim, A., Chaccour, K., El Hassani, A. and Andres, E.

SleepPal: A Sleep Monitoring System for Body Movement and Sleep Posture Detection.

DOI: 10.5220/0011840900003476

In Proceedings of the 9th International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AWE 2023), pages 40-47

ISBN: 978-989-758-645-3; ISSN: 2184-4984

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

considering the number of transitions between two

sleep postures over the period of the sleep.

The rest of this paper is divided as follows. Sec-

tion 2 surveys existing solutions related to bed move-

ments detection. Section 3 elaborates on the material

and methods followed in the development and imple-

mentation of the sleep monitoring system. Section 4

explains the experimental evaluation of the proposed

system and discusses the obtained results. Finally,

Section 5 presents the concluding remarks and the fu-

ture directions of this research.

2 RELATED WORK

Actigraphy has been used in various areas of medi-

cal research, typically for monitoring motion-related

sleep disorders. In fact, It has been used in many stud-

ies since 30 years ago. (Kupfer et al., 1972) reported

a correlation between EEG signals, wrist activity, and

wakefulness in 1972. (Sadeh et al., 1995) concluded

that normal subjects showed more than 90% correla-

tion when comparing actigraphy data with PSG. By

1995, sufﬁcient experimentation had been carried out

to ﬁnally enable the Standards of Practice Committee

of the American Sleep Disorders Association (ASDA)

to support the use of actigraphy in evaluating cer-

tain aspects of sleep disorders, such as insomnia, cir-

cadian sleep–wake disturbances, and periodic limb

movements. On an higher level, a generic classiﬁca-

tion of bed related systems was proposed by (Ibrahim

et al., 2021). The authors classiﬁed these systems

into Wearable Systems (WS), Non-Wearable Systems

(NWS), and Fusion Systems (FS).

• Wearable Systems (WS): Many studies have

been reported to use actigraphy in WS. For in-

stance, (Miwa et al., 2007) proposed a rollover

detection system using a SenseWear Pro2 Arm-

band. They used the maximum and the mean

acceleration in the x and y directions to identify

the posture differences using a threshold based

algorithm. The experiments showed that 82.4%

of rollovers were detected. Similarly, (Acharya,

2020) proposed a rollover detection system using

the ADXL335 accelerometer attached on socks

where acceleration row data was used in a thresh-

old based algorithm.

• Non-Wearable Systems (NWS): Other studies

reported the use of unobtrusive sensors used for

bed movement detection. These include load sen-

sors installed under bedposts. (Adami et al., 2008)

proposed a system for classiﬁcation of movement

in bed using load cells. The experiments were

done on 15 participants and showed an accuracy

of 84.6%. (Beattie et al., 2011) also used load

cells under bedposts giving a total of 4 cells. The

classiﬁcation of the sleeping posture was done us-

ing K-Means classiﬁer of the bed’s Center of Pres-

sure (CoP) in the x and y directions. The exper-

iments showed an accuracy of 68%, 57%, 69%,

and 33% for the back, right, left, and stomach

postures respectively. The problem with this ap-

proach is that CoPy values for back and stom-

ach postures are the same so the classiﬁer fails to

distinguish between them. An unrestrained sleep

monitoring system using cameras has been also

proposed to monitor sleep postures. (Lee et al.,

2015) proposed a system to monitor the sleep-

ing position using a Kinect sensor. They used

Kinect’s skeleton and Kinect’s infrared camera to

detect the body joints. The joint model given by

the Kinect has 25 points. The authors recorded the

x, y and z positions of those points and calculated

the sleep movement taking the Euclidean distance

between those points with respect to time. How-

ever, the procedure may still be considered as an

invasion of privacy for some patients which makes

its utilisation not suitable.

• Fusion Systems (FS): Fusion of sensors will form

a multi-channel source of data which can provide

more accurate analysis. For instance, (Nam et al.,

2016a) used a pressure sensor and an accelerom-

eter to extract data on motion, respiration, body

activity, and heart rate. These data were used to

measure sleep quality by estimating the depth of

sleep, the number of apneaic episodes and the pe-

riodicity. The experimental results demonstrated

that the proposed system is effective in measuring

the physiological factors of sleep quality.

In addition to proposing a low cost device used to

identify body posture and transitions, this paper seeks

to contribute in proposing a body transition matrix

that could be used in any threshold based algorithm.

3 MATERIALS AND METHODS

3.1 Wearable Sensor

Continuous sleep monitoring must be unobtrusive

and have minimal physical impacts on bed activi-

ties. For these reasons, we proposed our sleep mon-

itoring system SleepPal. SleepPal includes a small

device that could be attached to the center of the

chest and measures the body acceleration in the three

axes. The device transmits real-time accelerome-

SleepPal: A Sleep Monitoring System for Body Movement and Sleep Posture Detection

ter raw data over the Wi-Fi network to a monitor-

ing and visualization software installed on a remote

computer. The architecture of SleepPal is illustrated

in Figure 1. The system is composed of an Iner-

tial Measurement Unit (IMU) and a Microcontroller

Unit (MCU). To ensure the wireless communication,

ESP8266 NodeMCU is used with IMU MPU6050

which includes the ADXL345 triaxial accelerometer.

Figure 1: Architecture of SleepPal.

3.2 Sleep Postures and Bed Movements

The objective of SleepPal is to detect and record the

sleep posture and bed movements in order to build a

sleep monitoring database. The device will be able to

detect and store each posture and each movement in

the database for further analysis and evaluate the sleep

quality. Generally, sleep postures on the bed could be

classiﬁed into four categories namely, the front, the

back, the left, and the right (Nam et al., 2016b). The

same terminology is also adopted by (Hoque et al.,

2010). This terminology will be adopted for the rest

of this paper. These postures are as follows:

• Supine Posture (SP): The subject is lying on his

back (back posture);

• Prone Posture (PP): The subject is lying on his

stomach (front posture);

• Right Lateral Posture (RLP): The subject is lying

on his right side (right posture);

• Left Lateral Posture (LLP): The subject is lying

on his left side (left posture);

The sleeping posture is identiﬁed based on the

body movement detected and the raw data extracted

from the accelerometer. By knowing which move-

ment was performed, we can derive the position rela-

tively to the edges of the bed. The motion data is used

to validate the last performed posture.

A change in the sleeping posture also called

rollover is deﬁned as a series of trunk movements

beginning from the current static posture to the next

static posture through rotational motions during sleep

(Miwa et al., 2007) (Acharya, 2020). Therefore, the

most adequate location to ﬁx our device is on the

Figure 2: Common sleeping postures. From left to right:

Left Lateral Posture (LLP), Right Lateral Posture (RLP),

Prone Posture (PP) and Supine Posture (SP).

chest. Referring to this deﬁnition, each static posture

is deﬁned as a state and each rollover is deﬁned as a

transition. SleepPal will count the rollovers to deter-

mine the position and evaluate the sleep quality. We

identify four states and eight possible transitions as

illustrated in Figure 3.

Figure 3: Sleep postures and movement transition scheme.

On the side note, there can be other transitions

such that from RLP to LLP and vice versa or from

PP to SP and vice versa, however, such transitions are

a composition of two consecutive transitions.

3.3 Features Extraction

The data from the accelerometer is collected in raw

mode, which provides the acceleration data in actual

g-forces. The sampled data are stored and analysed in

epoch of an approximate length of 2s. The extracted

features are the Mean Acceleration (

(T )

(L)

), and

the Mean of Absolute Difference (MAD

(T )

, MAD

(L)

)

for both the transverse and longitudinal directions

as well as the Posture Difference (∆P) proposed by

(Miwa et al., 2007). Herein, MAD

(d)

is used as an

ICT4AWE 2023 - 9th International Conference on Information and Communication Technologies for Ageing Well and e-Health

indicator of the movement intensity, the Mean Accel-

eration

(d)

is used to identify the direction, and (∆P)

is used to identify the difference in postures. These

features are formulated below:

3.3.1 Mean of Acceleration ( ¯a)

It represents the average of acceleration in both trans-

verse and longitudinal directions (d) deﬁned in g and

shown in Equation (1).

(d)

∑

i=1

(1)

Where n is the number of samples in the epoch

and a

is the acceleration sample.

3.3.2 Mean of Absolute Difference (MAD)

It describes the mean distance of data points about the

mean in both transverse and longitudinal directions

(d) deﬁned in g and shown in Equation (2).

MAD

(d)

∑

i=1

− ¯r| (2)

Where n is the number of samples in the epoch, r

is the i

resultant sample within the epoch and ¯r is the

mean resultant value of the epoch.

3.3.3 Posture Difference (∆P)

It represents the difference in average acceleration

(T )

and

(L)

deﬁned in g

as shown in Equation (3).

We designated ∆P as Posture Difference between time

t and t − 1.

∆P

= (

(T )

−

(T )

t−1

)

+ (

(L)

−

(L)

t−1

)

(3)

Where

(T )

is the average acceleration in the

transverse direction, and

(L)

is the average acceler-

ation in the longitudinal direction.

3.4 Detection Algorithm

According to the position of the accelerometer in-

stalled in the wearable device illustrated in Figure 4,

the y-axis will represent the transverse direction and

the z-axis will represent the longitudinal direction.

When stationary, the acceleration over the z-axis is

equal to the gravitational force (+1g), and it should

be (0g) on the y-axis. Thus, the value of the z-axis

should be positive in SP, and negative in PP. On the

other hand, the value of the y-axis should be posi-

tive when moving right and negative when moving

left. Ideally, when moving to the LLP, the acceler-

ation over the y-axis should be equal to (−1g) and

(0g) in the z-axis. Alternately, when moving to the

RLP, the acceleration over the z-axis should be equal

to (+1g) and (0g) in the y-axis.

Figure 4: Positioning of SleepPal wearbale device.

The signal of ∆P and MAD

(d)

in both directions

during a transition or a rollover corresponds to an in-

crease in both values since MAD

(d)

is inﬂuenced by

the movement intensity and ∆P expresses the differ-

ence in posture between time t and t − 1. Signals of

the above features were exploited experimentally in

order to deﬁne and derive the below threshold values:

• Upper Posture value (U

∆P

): corresponds to the

lowest upper peak value of ∆P recorded.

• Upper MAD

(T )

Transversal value (U

MAD

(T )

): cor-

responds to the lowest upper peak value of

MAD

(T )

recorded.

• Upper MAD

(L)

Longitudinal value (U

MAD

(L)

): cor-

responds to the lowest upper peak value of

MAD

(L)

recorded.

The perfect lateral position is hard to achieve

therefore we deﬁned two additional thresholds:

• Upper

(L)

in both lateral postures (U

(L)

): corre-

sponds to the highest value of

(L)

in the lateral

position.

• Lowest

(L)

in both lateral postures (L

(L)

): cor-

responds to the Lowest value of

(L)

in the lateral

posture.

Thus, a transition is detected when ∆P, MAD

(T )

and MAD

(L)

are higher than the deﬁned thresholds as

described in Equation (4):







∆P

> U

∆P

MAD

(T )

> U

MAD

(T )

MAD

(L)

> U

MAD

(L)

(4)

To classify the transition upon detecting a move-

ment, we used

(T )

(L)

, and

(L)

t−1

. The pattern in

each transition is described in Table 1.

The proposed algorithm will use the peaks of the

posture difference ∆P, MAD

(T )

, and MAD

(L)

to de-

tect body transitions and rollovers. Upon detecting a

SleepPal: A Sleep Monitoring System for Body Movement and Sleep Posture Detection

Table 1: Body transition matrix.

↱

SP PP LLP RLP

NA NA

(L)

↘

(T )

↘

(L)

t−1

> U

(L)

↘

(T )

↗

(L)

t−1

> U

(L)

NA NA

(L)

↗

(T )

↘

(L)

t−1

< L

(L)

↗

(T )

↗

(L)

t−1

< L

(L)

LLP

(L)

↗

(T )

↗

(L)

t−1

< U

(L)

↘

(T )

↗

(L)

t−1

< U

(L)

NA NA

RLP

(L)

↗

(T )

↘

(L)

t−1

< U

(L)

↘

(T )

↘

(L)

t−1

< U

(L)

NA NA

body movement, the algorithm will compute the dif-

ference in signals for

(T )

and

(L)

in order to deter-

mine the signal behavior. Finally, the algorithm will

compare the signal patterns with the matrix of pattern

behaviors presented in Table 1 in order to identify the

transition type and the current sleep posture.

4 EXPERIMENTAL EVALUATION

The system was evaluated on data collected in labora-

tory. Experiments were conducted on 10 volunteering

subjects aging between 18 and 40 and weighting from

52 to 106kg. The device was attached to the center of

the chest. The volunteers didn’t report any discom-

fort in wearing the device. Each subject simulated all

sleep movements ﬁve times. Thus, each subject per-

formed 40 transitions and each posture transition is

performed 50 times. In order to conﬁrm the volun-

teer’s posture, we installed a video camera above the

bed and recorded the body movements.

Figures 5, 6, 7, and 8 show the signals recorded

during body movements transitioning contrariwise

from SP to LLP, from SP to RLP, from PP to LLP, and

from PP to RLP respectively. As shown, during any

body movement, a peak of ∆P, MAD

(T )

, and MAD

(L)

are recorded simultaneously. The signals of

(T )

and

(L)

differ according to the transition.

The signals of the

(L)

and

(T )

directions cor-

respond to a decrease in both values during SP to

LLP and RLP to PP transitions. This decrease is due

to rotational motion to the left. In order to distin-

guish between these transitions, we used the

(L)

t−1

identify the previous posture. Similarly, from the SP

to LLP transition, the

(L)

t−1

value should be (+1g)

which is greater than U

(L)

recorded in lateral pos-

tures, whereas in the RLP to PP transition, the

(L)

t−1

should be between L

(L)

and U

(L)

. Contrarily, the sig-

nals observed in LLP to SP and PP to RLP correspond

to an increase in both values but in the ﬁrst transition,

the

(L)

t−1

should be between L

(L)

and U

(L)

and in

the second transition it should be less than L

(L)

. An

increase in

(L)

along with a decrease in

(T )

are ob-

served in RLP to SP and PP to LLP transitions.

(L)

t−1

should be between L

(L)

and U

(L)

in the ﬁrst transition

and less than L

(L)

in the second. Finally, a decrease

(L)

with an increase in

(T )

are observed in SP

to RLP and LLP to PP transitions.

(L)

t−1

should be

greater than U

(L)

in the ﬁrst transition and between

(L)

and U

(L)

in the second.

Table 2 shows the detection results of the proposed

algorithm. Since we considered any body movement

is a rollover, the total number of rollovers is 400.

The proposed algorithm detected a 352 rollovers con-

ﬁrmed to be correct. Consequently, 48 rollovers were

undetected (false negative) and a total of 30 false pos-

itive detections were recorded. The efﬁcacy and ro-

bustness of our algorithm are evaluated by measur-

ing both the sensitivity (i.e. Detected rollovers over

the total of number of rollovers) and the accuracy (i.e.

Ability of the system to detect the rollovers) which are

recorded to be equal to 88% and 82% respectively.

ICT4AWE 2023 - 9th International Conference on Information and Communication Technologies for Ageing Well and e-Health

Figure 5: Signals during movement from SP to LLP and vice versa.

Figure 6: Signals during movement from SP to RLP and vice versa.

Figure 7: Signals during movement from PP to LLP and vice versa.

5 SLEEP QUALITY EVALUATION

Human sleep can be classiﬁed into Rapid Eye

Movement (REM) and Non-Rapid Eye Movement

(NREM). The latter is further divided into three

stages, N1-N3. Each stage of sleep includes varia-

tions in the brain wave pattern and eye movements.

The body cycles through all of these stages approxi-

mately 4 to 6 times each night, averaging 90 minutes

for each cycle (Memar and Faradji, 2017). REM and

N1 stages are the lightest sleep, while N3 stage is the

deepest sleep. In general, body movements increase

during light sleep and decrease during deep sleep.

Thus, the sleep can be classiﬁed into two stages, deep

sleep and light sleep based on the frequency of body

transitions detected. Hence, and upon detecting a

body transition, SleepPal will record the time and will

compare it to the time of the previously recorded body

transition. Consequently, the interval between body

transitions is computed and compared to a threshold

SleepPal: A Sleep Monitoring System for Body Movement and Sleep Posture Detection

Figure 8: Signals during movement from PP to RLP and vice versa.

Table 2: Results of the detection algorithm.

# Posture Transitions Detection

1 SP → LLP 46/50

2 SP → RLP 47/50

3 LLP → SP 45/50

4 RLP → SP 45/50

5 PP → LLP 43/50

6 PP → RLP 43/50

7 LLP → PP 42/50

8 RLP → PP 41/50

as shown in Equation (5).



DeepSleep = BM

− BM

t−1

≥ time

LightSleep = BM

− BM

t−1

< time

(5)

Where BM

and BM

t−1

are the body movements

at times t and t − 1 respectively. time

is the time

threshold equal to 20 minutes according to (Miwa

et al., 2007).

The time of each sleep stage will be recorded in

order to evaluate the sleep quality. The latter is de-

pending on several factors such as genetics, sleep

habits, medical problems, and essentially sleep depth

which is considered the most important in evaluating

sleep quality. Thus, we determined the sleep quality

in term of the sleep depth according to Equation (6)

as follows:

SleepQuality

DeepSleep

Total

(6)

Where QI

SleepQuality

is the sleep Quality Index,

DeepSleep

is the duration of the deep sleep stage, and

Total

is the total sleep duration.

6 CONCLUSION

In this paper, we developed a body movement and

body posture classiﬁer using an accelerometer device

attached to the center of the chest. The proposed sys-

tem is based on a simple threshold algorithm and uses

multiple features extracted from the raw acceleration

data. The efﬁcacy and robustness of our algorithm are

evaluated by measuring both the sensitivity and the

accuracy which were recorded to be equal to 88% and

82% respectively. In order to evaluate the sleep qual-

ity, we distinguished between deep sleep and light

sleep using a simple threshold-based equation. Our

proposed sleep monitoring system can be used for

monitoring sleep quality in hospitals by interfacing it

with the existing nurse call system.

Further research will exploit the limitations of the

device in settings where sliding in bed remains un-

detected and motion noise caused by the ﬁxation of

the sensor must be removed. We will also emphasize

the relationship between sleep quality, sleep disorders

and nocturnal falls.

REFERENCES

Acharya, J. (2020). Design and development of a method

for detecting sleep roll-over counts using accelerom-

eter adxl335. International Journal of Electrical and

Computer Engineering (IJECE), 10:477.

Adami, A., Singer, C., Hayes, T., and Pavel, M. (2008).

A system for unobtrusive monitoring of mobility in

bed. Proceedings of the 11th IEEE International Con-

ference on Computational Science and Engineering,

CSE Workshops 2008, pages 13 – 18.

Beattie, Z., Hagen, C., and Hayes, T. (2011). Classiﬁcation

of lying position using load cells under the bed. Con-

ference proceedings : ... Annual International Confer-

ence of the IEEE Engineering in Medicine and Biol-

ICT4AWE 2023 - 9th International Conference on Information and Communication Technologies for Ageing Well and e-Health

ogy Society. IEEE Engineering in Medicine and Biol-

ogy Society. Conference, 2011:474–7.

Cartwright, R. (1984). Effect of sleep position on sleep ap-

nea severity. Sleep, 7:110–4.

Colman, A. (2006). A Dictionary of Psychology. Oxford

University Press.

De Koninck, J., Gagnon, P., and Lallier, S. (1983). Sleep

positions in the young adult and their relationship with

the subjective quality of sleep. SLEEP, 6:52–59.

Hoffstein, V. (1996). Snoring. chest, 109:201–22.

Hoque, E., Dickerson, R., and Stankovic, J. (2010). Mon-

itoring body positions and movements during sleep

using wisps. Proceedings - Wireless Health 2010,

WH’10, pages 44–53.

Horne, R., Franco, P., Adamson, T., Groswasser, J., and

Kahn, A. (2002). Effects of body position on sleep

and arousal characteristics in infants. Early human

development, 69:25–33.

Ibrahim, A., Chaccour, K., Hassani, A. H. E., and An-

dres, E. (2021). Bed-fall detection and prediction:

A generic classiﬁcation and review of bed-fall related

systems. IEEE Sensors Journal, 21(5):5678–5686.

Kupfer, D., Detre, T., Foster, G., Tucker, G., and Delgado,

J. (1972). The application of delgado’s telemetric mo-

bility recorder for human studies. Behavioral biology,

7:585–90.

Lee, J., Hong, M., and Ryu, S. (2015). Sleep monitoring

system using kinect sensor. International Journal of

Distributed Sensor Networks, 2015:1–9.

Leung, R., Bowman, M., Parker, J., Newton, G., and

Bradley, T. (2003). Avoidance of the left lateral decu-

bitus position during sleep in patients with heart fail-

ure: Relationship to cardiac size and function. Journal

of the American College of Cardiology, 41:227–30.

Memar, P. and Faradji, F. (2017). A novel multi-class eeg-

based sleep stage classiﬁcation system. IEEE Trans-

actions on Neural Systems and Rehabilitation Engi-

neering, PP:1–1.

Miwa, H., Sasahara, S., and Matsui, T. (2007). Roll-

over detection and sleep quality measurement using

a wearable sensor. Conference proceedings : ... An-

nual International Conference of the IEEE Engineer-

ing in Medicine and Biology Society. IEEE Engi-

neering in Medicine and Biology Society. Conference,

2007:1507–10.

Nam, Y., Kim, Y., and Lee, J. (2016a). Sleep monitoring

based on a tri-axial accelerometer and a pressure sen-

sor. Sensors (Basel, Switzerland), 16.

Nam, Y., Kim, Y., and Lee, J. (2016b). Sleep monitoring

based on a tri-axial accelerometer and a pressure sen-

sor. Sensors, 16:750.

Sadeh, A., Hauri, P., Kripke, D., and Lavie, P. (1995). The

role of actigraphy in the evaluation of sleep disorders.

SLEEP, 18:288–302.

SleepPal: A Sleep Monitoring System for Body Movement and Sleep Posture Detection