Increased Fall Detection Accuracy in an Accelerometer-based Algorithm

Considering Residual Movement

Panagiotis Kostopoulos

, Tiago Nunes

, Kevin Salvi

, Michel Deriaz

and Julien Torrent

Institute of Services Science, University of Geneva, Geneva, Switzerland

Fondation Suisse pour les T´el´eth`eses, Neuchˆatel, Switzerland

Keywords:

Fall Detection, Smartwatch, Sensors, Residual Movement, Accelerometer, Alarm.

Abstract:

Every year over 11 million falls are registered. Falls play a critical role in the deterioration of the health of the

elderly and the subsequent need of care. This paper presents a fall detection system running on a smartwatch

(F2D). Data from the accelerometer is collected, passing through an adaptive threshold-based algorithm which

detects patterns corresponding to a fall. A decision module takes into account the residual movement of

the user, matching a detected fall pattern to an actual fall. Unlike traditional systems which require a base

station and an alarm central, F2D works completely independently. To the best of our knowledge, this is the

ﬁrst fall detection system which works on a smartwatch, being less stigmatizing for the end user. The fall

detection algorithm has been tested by Fondation Suisse pour les T´el´eth`eses (FST), the project partner for the

commercialization of our system. Taking advantage of their experience with the end users, we are conﬁdent

that F2D meets the demands of a reliable and easily extensible system. This paper highlights the innovative

algorithm which takes into account residual movement to increase the fall detection accuracy and summarizes

the architecture and the implementation of the fall detection system.

1 INTRODUCTION

Unintentional falls are frequent and quite dangerous

for elderly people. Yearly, more than 11 million falls

are registered (Brown, 2005), leading to a wide spec-

trum of injuries for this age group. Aside from caus-

ing physical injuries, falls can also have dramatic psy-

chological consequences that reduce elderly people’s

independence (Ryynanen et al., 1992). It has been

found that after falling, 48% of older people report

a fear of falling and 25% report curtailing activities.

Moreover, falls can also lead to disability and de-

creased mobility which often results in increased de-

pendency on others and, hence, an increased need of

being admitted to an institution. Finally, one other se-

rious consequence of falling is the ”long-lie” condi-

tion where a falling person remains on the ground or

ﬂoor for more than an hour after a fall. The ”long-lie”

is a sign of weakness, illness and social isolation and

is associated with high mortality rates among the el-

derly. Time spent on the ﬂoor can be associated with

a fear of falling, muscle damage, pneumonia, pres-

sure sores, dehydration and hypothermia (Lord et al.,

2001; Nevit et al., 1989).

In an attempt to minimize these serious conse-

quences of falling, various fall detection systems

were developed over the last decade. These systems

are mainly based on video-cameras (Nait-Charif and

McKenna, 2004; Lee and Lee, 2009; Huang et al.,

2008), acoustic (Alwan et al., 2006; Litvak et al.,

2008) or inertial sensors (Hwang et al., 2004) and

mobile phone technology (Sposaro and Tyson, 2009;

Dai et al., 2010; Dumitrache and Pasca, 2013; Aguiar

et al., 2014).

Common fall detection systems are based on

a sensor detecting a strong vertical acceleration,

launching an alarm when a fall event is recognized.

More recent systems usually take into account other

sensors able to detect the device’s orientation in or-

der to determine whether the user is lying or standing.

In (Dai et al., 2010) the authors present PerFallD, a

system which combines the detection and the com-

munication components using mobile phones. They

compare it with existing academic and commercial

solutions and conclude that their system is quite ac-

curate despite the fact that they are using only low-

cost sensors (the ones from an Android G1 phone). In

(Sposaro and Tyson, 2009) the authors present iFall,

another Android application tested on a G1 phone,

which is focused on the communication between the

Kostopoulos P., Nunes T., Salvi K., Deriaz M. and Torrent J..

Increased Fall Detection Accuracy in an Accelerometer-based Algorithm Considering Residual Movement.

DOI: 10.5220/0005179100300036

In Proceedings of the International Conference on Pattern Recognition Applications and Methods (ICPRAM-2015), pages 30-36

ISBN: 978-989-758-077-2

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

fall detection system and the alarm notiﬁcation mod-

ule, instead of the sensors only.

Most of the current fall detection systems require

a base central. In this paper we propose a fall de-

tection system (F2D) which works on a smartwatch,

therefore completely independent from a base station.

Using such a device is less stigmatizing for the user.

In addition, it can be offered for less than half of the

cost of existing systems on the market. Our system

meets the requirements of reliability, ease of installa-

tion and restriction of false positives (Doughty et al.,

2000) which are essential for a properly built fall de-

tection system.

F2D works on a smartwatch and therefore ﬁxed

on the wrist of the person. We have avoided the dis-

advantages of (He et al., 2012) where the solution of

the waist-mounted smartphone the authors provide is

not feasible for two reasons: 1) Normally people do

not wear their phones on the waist but in their pockets.

2) The system will be working only when the smart-

phone is mounted on the waist and not at other times.

Other problems such as the usage of intrusive devices

exist in (Hou et al., 2012) and (Li et al., 2012), where

the accelerometer and Bluetooth unit are bounded as

a wearable unit and placed on the subject’s waist or

chest.

Nowadays, simple smartwatches are very power-

ful and have a set of sensors that can be used and

diverted from their original intent. More computing

power and storage on these devices offer greater op-

portunities. Using a single smartwatch as a device for

running the F2D application satisﬁes the condition of

ease of installation of the fall detection system.

In general the fall detection algorithms wait for

a response of the user after a possible fall in order

to make a decision if the situation is critical or not.

But this is not always applicable and convenient for

the end user, especially for elderly people who are the

target group of our research. In F2D we propose an

accelerometer-based algorithm considering the resid-

ual movement after the fall. This analysis is per-

formed in the decision module of the F2D application

which is responsible for the classiﬁcation of a possi-

ble fall pattern being a real fall event or not.

The rest of this paper is organized as follows. In

Section 2 our designed fall detection system is de-

scribed in detail emphasizing the innovative fall de-

tection algorithm. Experimental results are reported

and discussed in Section 3. Future improvements of

our work to make F2D more robust are presented in

Section 4. Finally, a brief conclusion is given in Sec-

tion 5.

2 SYSTEM DESIGN

Our fall detection system is an Android application

running on an AW-420.RX smartwatch of Simvalley

Mobile. We have chosen the Android based solution

because it is an open source framework designed for

mobile devices. The Android SDK provides the API

libraries and developer tools necessary to build, test

and debug applications for Android. We implemented

the prototype in Java using the Android SDK API 19.

The fall detection algorithm, which is explained be-

low, is implemented as a background service. When a

fall is detected the service informs the main applica-

tion, which notiﬁes the caretakers (family or friends).

There are several notiﬁcation channels the user can

choose from: call, SMS or email.

2.1 System Overview

We collect data from the accelerometer sensor of the

smartwatch. Then we apply ﬁlters to detect patterns

corresponding to a fall. We use a threshold based

algorithm for the fall detection which takes into ac-

count the residual movement of the user after the fall.

The thresholds were selected based on experiments

with different proﬁles of users (age, weight, height

are factors that were taken into account). The deci-

sion module combines different data coming from the

ﬁlters as depicted in Figure 1, in order to make a deci-

sion whether a possible fall corresponds to a real fall.

If it is the case, the information is transmitted to the

alarm module. When a potentially critical situation is

detected, the smartwatch uses different communica-

tion means (call, SMS, email) to inform the caretak-

ers. Another important difference with the traditional

systems is that the smartwatch communicates directly

with the caretakers with no involvement of a base sta-

tion and a centralized alarm.

Figure 1: System architecture.

2.2 Fall Detection Algorithm

The fall detection algorithm is implemented in a back-

ground service and is running continuously. The user

can operate his smartwatch as usual. F2D does not

cause any interference with the normal usage of in-

stalled applications. The algorithm is threshold based

like (Dumitrache and Pasca, 2013), relying on the

captured data of the accelerometer of the smartwatch.

IncreasedFallDetectionAccuracyinanAccelerometer-basedAlgorithmConsideringResidualMovement

We decided to use a threshold based algorithm and

not a machine learning approach like (Aguiar et al.,

2014) as it is less complex and therefore requires the

lowest computational power (Habib et al., 2014). In

the typical scenario, the user will use the application

on his smartwatch normally during the day without

the requirement of charging it much more than usu-

ally. Since the fall detection system will run continu-

ously, we should optimize the battery consumption of

the device. Therefore, only the tri-axial accelerometer

signal is used since it is the most informative sensor

regarding the fall detection.

The algorithm distinguishes daily activities from

falls. Activities of Daily Living (ADL) are normal

activities such as walking, standing or running. The

pattern of a fall must be different from the patterns

of these activities. Acceleration data is sampled at 40

Hz from the 3-axis accelerometer sensor embedded

in the Android smartwatch. Speciﬁcally, the sensor

which provides acceleration information without the

gravity component (linear acceleration) is used. We

calculate the norm of the acceleration for each mo-

ment as described in Equation 1.

acceleration =

+ y

+ z

(1)

We have analyzed a set of data with 150 different

simulated falls from different people involved in the

experiments, as reported in Table 2, from our project

partner FST. This company has a long experience

in creating and using innovative products adapted to

people with disabilities. Thanks to this data we have

improved the detection of possible falls. We observed

that all falls follow one of the three patterns which we

have called smooth, strong and sharp. They are given

in Figures 2, 3 and 4 respectively. The main differ-

ence between them is the time interval of the residual

movement after the fall Table 1. More speciﬁcally,

when a fall takes place, the peak that exceeds the up-

per threshold of the acceleration corresponds to the

hit. After this, the pattern of the fall has a second

peak, lower than the ﬁrst one and exceeding the lower

threshold. Finally, the acceleration returns to normal

values. This behaviour of the acceleration after the

ﬁrst peak represents the residual movement that we

take into account in the decision module for the char-

acterization of a possible fall event as a real fall.

2.2.1 Time Window

The time window is an essential part of the fall de-

tection algorithm. We have deﬁned a time window in

which we are able to recognize a fall pattern. This

window is set to 6 seconds, a value which has been

selected after conducting experiments, using the set

of data from simulated falls mentioned above. The

main goal of the algorithm is the detection of all falls

and at the same time the elimination of false positives.

Building and testing our system we concluded that

less than 6 seconds is not enough for the detection of

all different types of falls. However, setting the win-

dow to higher values creates a bigger occurrence of

false positives.

2.2.2 Fall Pattern

The next step of the algorithm is the detection of a

possible fall. In order to consider an activity as a pos-

sible fall the two following conditions must be satis-

ﬁed: 1) If the acceleration exceeds an upper thresh-

old which is set to 18 m/sec

. 2) If after a given

time interval the acceleration exceeds a lower thresh-

old which is set to 7 m/sec

. The two thresholds have

been selected based on the basic trade-off between de-

tecting all falls and avoiding false positives. This time

difference between the two peaks represents the resid-

ual movement of the user after a fall. The intensity of

this movement depends on the proﬁle of the individ-

ual user. The three possible values of the time interval

are given in Table 1.

If the two conditions are satisﬁed during the time

window of 6 seconds then a possible fall is suspected.

We can see in Figures 2 - 4 that this time window

is sufﬁcient for the satisfaction of the two conditions

that should happen in order to detect a fall pattern.

2.2.3 Decision Module

The ﬁnal step of the fall detection algorithm is the

classiﬁcation of the fall pattern as real. In this deci-

sion module a counter increases every time that both

conditions of exceeding the thresholds are satisﬁed.

The critical range of the values of the fall counter

is (1 ≤ counter < 14). If (counter ≥ 14), then it is

due to another activity being performed (e.g., run-

ning) which gives the difference in the acceleration

values as we can see in Figure 6. On the other hand if

(counter < 1) it means the user at most did a sudden

movement with his wrist and so the threshold condi-

tions were not satisﬁed (e.g., when a user was going

down the stairs in Figure 8). The graphical explana-

tion and the structure of the fall detection algorithm is

given in Figure 5.

2.3 Emergency Actions

If the algorithm decides that a fall has happened then

the background service notiﬁes the main application,

which in turn sends a message to the caretakers.

ICPRAM2015-InternationalConferenceonPatternRecognitionApplicationsandMethods

The smartwatch communicates directly with the

caretakers. In case of an alarm the loudspeaker of

the watch is automatically turned on at a high volume

and calls from caretakers are automatically answered.

This allows the user to communicate even in uncom-

fortable positions that could result after a fall.

Table 1: The three possible time intervals.

Smooth Strong Sharp

100 ms 300 ms 500 ms

Figure 2: Smooth fall.

Figure 3: Strong fall.

3 EVALUATION

For the evaluation of the reliability of the fall detec-

tion mechanism in F2D we performed a series of ex-

periments. We collected different types of falls’ data

(i.e., forwards, backwards, sideways). We also col-

lected activities of daily living data (e.g., walking, go-

ing down the stairs, going up the stairs) as we can

see in Figures 7 - 9. Since it is very difﬁcult to test

a fall detection mechanism with elderly people, as it

can more easily cause injuries, we performed the ex-

Figure 4: Sharp fall.

Figure 5: Fall detection algorithm.

Figure 6: Running activity.

periments with healthy adults. The test subjects were

wearing smartwatches while falling on a mattress.

IncreasedFallDetectionAccuracyinanAccelerometer-basedAlgorithmConsideringResidualMovement

Figure 7: Walking activity.

Figure 8: Going down the stairs.

Figure 9: Going up the stairs.

For testing the ADL the test subjects wore the

smartwatches for 24 hours. Testers were doing all

their daily life activities. As we can see in the Fig-

ures 7 - 9 the acceleration values do not exceed the

upper threshold. Hence, a suspected fall is not classi-

ﬁed as a real fall because the counter must exceed 0

in order to detect a real fall.

Moreover, we evaluated our fall detection algo-

rithm using a set of simulated falls and another set

with ADL data, that we received from our project

partner FST. This set consists of a subset of 384 sim-

ulated falls and 417 ADL. The set of data that we

are using is much larger comparing with other sys-

tems (Dumitrache and Pasca, 2013) where only 34

simulated falls and 200 daily activities simulated by

a single young person were used. Our data has been

collected from 6 people with different proﬁles as re-

ported in Table 2. The simulated falls took place in a

room falling on a mattress in order to avoid injuries.

Also the ADL were the following: walking, going up

the stairs, going down the stairs, stand up from a chair,

sit down on a chair, running. Based on these facts,

it can be noted that the accuracy of our algorithm is

quite high. We achieved a true positive rate (sensi-

tivity) of 92.18% for the set of simulated falls and a

true negativerate (speciﬁcity) of 87.29% for detecting

the ADL data. The average of sensitivity and speci-

ﬁcity represents the accuracy of the system which is

equal to 89.74%. The analytical results are presented

in Figures 10 and 11.

Table 2: Different proﬁles.

Age Height (cm) Weight (kg)

22 185 76

26 176 69

27 182 63

29 184 53

30 186 93

40 177 75

Figure 10: Accuracy using real data from partner.

We conclude that the F2D system works reliably.

Some false alarms were detected when the testers per-

formed sudden movements with residual activity try-

ing to simulate the same pattern of a fall event.

4 FUTURE WORK

The ﬁrst step for further improvement of the robust-

ness of our fall detection system is the reduction of

ICPRAM2015-InternationalConferenceonPatternRecognitionApplicationsandMethods

Figure 11: Decomposition of ADL and respective detection

rate.

false positives. We are planning to achieve this goal

in three steps. The ﬁrst step is the elimination of mis-

leading direction of the movement of the smartwatch.

We cannot expect to detect a fall when the direction

of smartwatch is going up. Thanks to this recognition

of the movement direction we will avoid some false

positivescoming from sudden movementswhich have

similar patterns with a fall but in which the direction

of the movement is going up.

The second step for reducing false positives is the

use of adaptive thresholds. With the selected thresh-

olds we cover most of the cases of a fall based on the

experiments carried out in our lab and the data that

we received from the end users of our project partner.

With the adaptivity of the upper and lower thresholds

we will make our fall detection system more generic

and therefore allow a tailored response for each spe-

ciﬁc end user. We expect further improvements to the

fall detection as well as a reduction of the false posi-

tives.

The third and last step in decreasing false positives

is taking into account the users’ context. Common

fall detection systems are based on a sensor detect-

ing a strong vertical acceleration. But smoother falls

(e.g., a person grabbing a chair, a table, or any other

object) are not detected while some particular situa-

tions (a user that sits abruptly on his sofa) lead to false

alarms. To distinguish only the critical situations, we

will take into account contextual data. For example,

what happened before the fall, after the fall, at which

place in the room, at what time of the day, or next to

what object in a room. A user’s proﬁle, continuously

updated with his habits or particular behaviour, will

bring additional data to the user’s context. The cur-

rent context (e.g., position, time) will be used in two

different manners. The ﬁrst is to deﬁne the variables

of the ﬁlters in order to better differentiate the pos-

sible (expected) patterns. For example, we can take

into account the users’ height and weight as parame-

ters for the algorithm. The second manner is to help

the decision module. Once a possible fall is detected

by the algorithm, we will check if it is likely that it is

a real fall according to the current context.

Finally, thanks to the experience of our project

partner with elderly end users, we will be able to test

our fall detection system with real-world fall data in

the future. Testing our algorithm in real-life condi-

tions will give us the opportunity to create a more ro-

bust automated alarm system with higher acceptance.

5 CONCLUSION

Fall detection is a research ﬁeld that has a big impact

on the improvementof the daily life of elderly people.

In this paper we propose the ﬁrst fall detection system

designed to run on an independent smartwatch (F2D).

There is no base station (which limits the range), no

central alarm station (which is more difﬁcult to man-

age) and it works on a standard smartwatch. It implies

that it is less stigmatizing for the end user, quite cheap

comparing to existing systems and it is easily extend-

able. F2D uses an innovative fall detection algorithm

which takes into account the residual movementof the

user in order to match a fall pattern to a real fall.

We have conducted a range of experiments in our

lab and used real data from our project partner FST

who has relevant experience in the domain involving

real end users. These experiments demonstrated that

the fall detection system is robust.

Based on the reliability of the fall detection and

the restriction of false positives, which are guaran-

teed by the fall detection algorithm, we have built a

system which meets the requirements for deployment

and use.

ACKNOWLEDGEMENTS

This work was supported by the Swiss Commission

for Technology and Innovation (CTI grant 15876.2

PFES-ES). The authors would like to thank Anastasija

Collen who helped to improve the quality of this pa-

per.

REFERENCES

Aguiar, B., Rocha, T., Silva, J., and Sousa, I.

(2014). Accelerometer-based fall detection for smart-

phones. In Medical Measurements and Applications

(MeMeA), 2014 IEEE International Symposium on.

IncreasedFallDetectionAccuracyinanAccelerometer-basedAlgorithmConsideringResidualMovement

Alwan, M., Rajendran, P., Kell, S., Mack, D., and Dalal, S.

(2006). A smart and passive ﬂoor-vibration based fall

detector for elderly. In Proc. ICTTA’ 06.

Brown, G. (2005). An accelerometer based fall detec-

tor: Development, experimentation, and analysis. In

EECS/SUPERB.

Dai, J., Bai, X., Yang, Z., Shen, Z., and Xuan, D. (2010).

Perfalld: A pervasive fall detection system using mo-

bile phones. In Pervasive Computing and Communi-

cations Workshops (PERCOM Workshops), 2010 8th

IEEE International Conference on.

Doughty, K., Lewis, R., and McIntosh, A. (2000). The

desigh of a practical and reliable fall detector for

community and institutional telecare. In Journal of

Telemedicine and Telecare.

Dumitrache, M. and Pasca, S. (2013). Fall detection al-

gorithm based on triaxial accelerometer data. In E-

Health and Bioengineering Conference (EHB), 2013.

Habib, M. A., Mohktar, M. S., Kamaruzzaman, S. B.,

Lim, K. S., Pin, T. M., and Ibrahim, F. (2014).

Smartphone-based solutions for fall detection and

prevention: challenges and open issues. In

http://www.mdpi.com/journal/sensors.

He, Y., Li, Y., and Bao, S.-D. (2012). Fall detection by

built-in tri-accelerometer of smartphone. In Biomedi-

cal and Health Informatics (BHI), 2012 IEEE-EMBS

International Conference on.

Hou, Y., Li, N., and Huang, Z. (2012). Triaxial

accelerometer-based real time fall event detection. In

Information Society (i-Society), 2012 International

Conference on.

Huang, B., Tian, G., and Li, X. (2008). A method for fast

fall detection. In Proc 7th World Congr Intell Contr

Autom.

Hwang, J., Kang, J., Jang, Y., and Kim, H. (2004). Devel-

opment of novel algorithm and real-time monitoring

ambulatory system using bluetooth module for fall de-

tection in the elderly. In Proc 26th IEEE EMBS Ann

Int Conf.

Lee, Y. and Lee, H. (2009). Multiple object tracking for

fall detection in real-time surveillance system. In Proc

11th Int Conf Adv Commn Tec.

Li, Y., Chen, G., Shen, Y., Zhu, Y., and Cheng, Z. (2012).

Accelerometer-based fall detection sensor system for

the elderly. In Cloud Computing and Intelligent Sys-

tems (CCIS), 2012 IEEE 2nd International Confer-

ence on.

Litvak, D., Zigel, Y., and Gannot, I. (2008). Fall detection

of elderly through ﬂoor vibrations and sound. In Proc

30th IEEE EMBS Ann Int Conf.

Lord, R., Sherrington, C., and Menz, H. (2001). Falls in

older people: risk factors and strategies for prevention.

In Cambridge University Press. Cambridge.

Nait-Charif, H. and McKenna, S. (2004). Activity summari-

sation and fall detection in a supportive home environ-

ment. In Proc 17th Inte Conf Patt Recog. Cambridge

UK.

Nevit, M., Cummings, S., Kidd, S., and Black, D. (1989).

Risk factors for recurrent nonsyncopal fall. a prospec-

tive study. In J Am Med Ass.

Ryynanen, O. P., Kivela, S. L., Honkanen, R., and Laippala,

P. (1992). Falls and lying helpless in the elderly. In Z

Gerontol.

Sposaro, F. and Tyson, G. (2009). ifall: An android applica-

tion for fall monitoring and response. In Engineering

in Medicine and Biology Society, 2009. EMBC 2009.

Annual International Conference of the IEEE.

ICPRAM2015-InternationalConferenceonPatternRecognitionApplicationsandMethods