Pre-impact Fall Detection using Wearable Sensor Unit

Soonjae Ahn, Isu Shin, Bora Jeong and Youngho Kim

Department of Biomedical Engineering and Institute of Medical Engineering, Yonsei University, Wonju, South Korea

Keywords: Pre-impact Fall Detection, Accelerometer, Gyro Sensor, Vertical Angle.

Abstract: In this study, we verified our pre-impact fall detection algorithm through a clinical trials using wearable

sensor (accelerometer and gyro sensor) at waist. Forty male volunteers participated in the clinical trial (three

types of falls and seven types of ADLs). Results show that falls could be detected with an average lead-time

of 530ms before the impact occurs, with no false alarms (100% specificity) and no incorrect detects (100%

sensitivity). Our algorithm for pre-impact fall detection with a wearable sensor unit could be very helpful to

minimize fall risk.

1 INTRODUCTION

Falls are a major cause of injuries and deaths in

older adults (Annekenny and O’shea, 2002). Even

though most falls produce no serious injury, 20-30%

of fall related patients will suffer moderate to severe

injuries. Furthermore, some of them require

hospitalization to continue living in community and

have an increased risk of death (Nevitt et al., 1991;

Tinetti et al., 1995). Approximately 35% of

community-dwelling older adults and 50% of older

adults residing in long-term care facilities fall at

least once per year. The development of system to

prevent falls and fall-related injuries in older adults

is a major public health priority.

The most promising fall prevention strategy

involves the identification of individuals who had

increased fall risk and the implementation of the

appropriate prevention mechanism. Furthermore, it

includes physical restraint (Gross et al., 1990),

investigation of fall-related fractures prevention

strategies (Smeesters et al., 2001; Van den

Kroonenberg et al., 1996; Yamamoto et al., 2006),

study of characteristics and risk factors of syncope

(Kenny and O’Shea, 2002; Peczalski et al., 2006),

and multi-factorial risk assessment and management

(Weatherall, 2004).

As for intervention strategies, one of the

important problem in preventing or reducing the

severity of injury in the elderly is to detect falls in its

descending phase before the impact (pre-impact fall

detection) (Hayes et al., 1996). A few groups have

attempted to detect falls before impact (Bourke et al,

2008; Nyan at al., 2006; Wu, 2000). Wu

implemented pre-impact fall detection algorithm

using threshold of the horizontal and vertical

velocity profiles of the trunk using motion analysis

system. He showed that falls can be distinguished

from activities of daily living (ADL) with 300–

400ms lead-time before the impact (Wu, 2000).

Bourke et al. investigated pre-impact detection

algorithm of falls using threshold of the vertical

velocity of the trunk (Bourke et al., 2008). An

optical motion capture system and an inertial sensor

consisting of a tri-axial accelerometer and a tri-axial

gyroscope were used in their experiments. The

inertial sensor was located on the chest of the body.

Falls can be distinguished from ADLs, with 100%

accuracy and with an average of 323ms prior to

trunk impact and 140ms prior to knee impact, in that

subject group (Bourke et al., 2008). In pre-impact

fall detection, if a fall can be detected in its earliest

stage in the descent phase, more efficient impact

reduction systems can be implemented with a longer

lead-time for injury minimization (Hayes et al.,

1996).

In this study, we implemented a pre-impact fall

detection algorithm using a wearable sensor

positioned at waist. To verify our pre-impact fall

detection algorithm, three types of falls and seven

types of ADLs were conducted based on the

characteristics of angular movements of the sensor.

207

Ahn S., Shin I., Jeong B. and Kim Y..

Pre-impact Fall Detection using Wearable Sensor Unit.

DOI: 10.5220/0004902602070211

In Proceedings of the International Conference on Biomedical Electronics and Devices (BIODEVICES-2014), pages 207-211

ISBN: 978-989-758-013-0

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

2 MATERIALS AND METHODS

2.1 Subjects and Experiments

Forty healthy male volunteers participated in the

present study. Subject information in the study was

shown in Table 1.

Table 1: Subject information in the study.

n = 40

Subject information Mean ± SD

Age (year)

23.4 ± 4.4

Weight (kg)

68.7 ± 8.9

Height (cm)

172.0 ± 7.1

The experimental protocol was approved by the

Yonsei University Research Ethics Committee

(1041849-201308-BM-001-01) and written informed

consent was obtained from each subject. In faint

falls simulations, the subjects were told to stand on

the floor beside the mattress. Then they fell by

simply relaxing to the side, back, and front. All falls

were conducted on soft foam mattress for five times

respectively. A chair and the mattress were used for

the ADL trials (sitting, sit–stand transitions,

walking, stand–sit transitions, lying, jump, running).

Each activity was conducted for three times. The

algorithm was determined using experimental data

of twenty subjects, and then verified with blind test

data of twenty subjects. All clinical trials (falls and

ADLs) were recorded by a Bonita camera (Vicon

Motion Systems Ltd, UK) at frame rates of 200

frames/s.

MPU-9150 (Invensens®, USA) containing a 3-

axis accelerometer and a 3-axis gyro sensor was

used for the pre-impact fall detection sensor. The

definition of the sensor axis are shown in Figure 1.

The sensor was attached on the middle of the left

and the right anterior superior iliac spines. Data was

sampled at 100Hz.

Figure 1: Definition of the sensor axis.

2.2 Pre-impact Fall Detection

Algorithm

The algorithm was applied to falls and ADLs for

twenty subjects. For rapid detection before the

impact, threshold of acceleration and angular

velocity was set to 0.8g and 30°/s, respectively.

Furthermore, the threshold of vertical angle was set

to 30° because the maximum angle in the ADL does

not exceed beyond 30°, and we confirmed that the

angle during the ADL was not over 30° (Figure 2).

Lead time was defined as the time between impact

and detection (Figure 3). The process flow of pre-

impact detection algorithm in the processing unit is

shown in Figure 4. Acceleration data was

transformed into angles in sagittal and lateral planes,

measuring how many degrees these body segments

deviate from the vertical axis (i.e., standing is 0° and

supine on the floor is 90°), using the following

equations:

Deg

SAG

= tan

-1

acc

/ Y

acc

)*(180/π) (1)

and

Deg

LAT

= tan

-1

acc

/ Y

acc

)*(180/π) (2)

If sum of acceleration vectors is less than 0.8g and

angular velocities (|ω

SAG

|, |ω

LAT

|) are larger than the

30°/sec and vertical angles (|Deg

SAG

|Deg

LAT

|) larger

than 30° threshold level, the sensor detects a fall.

Figure 2: Acceleration, angular velocity and angular data

for stand-sit activity.

BIODEVICES2014-InternationalConferenceonBiomedicalElectronicsandDevices

208

Figure 3: Acceleration, angular velocity and angular data

for backward fall.

Figure 4: Pre-impact fall detection algorithm.

3 RESULTS

Table 2 showed peak acceleration, angular velocity

and angle during falls and ADLs. The results

showed that both acceleration and angular velocity

were greater than the threshold during several ADLs

while the angle did not exceed the threshold. For the

angle, it exceeded the threshold during sit-lying

activity only, but acceleration did not reach the

threshold during sit-lying activity. The algorithm

verified with blind test for twenty subjects. In the

blind test, no false detects was found in the

experiment (100% sensitivity) for all falls.

Furthermore, no incorrect detection was found in the

experiment (100% specificity) for all ADLs. Means

and standard deviations of lead times for the three

types of falls were shown in Figure 5. The lead time

was 474 ± 38.3ms, 590.3 ± 122.6ms and 527 ±

62.3ms in the backward, the forward and the side

falls respectively in order.

Figure 5: Mean and standard deviations of lead time in

backward, forward and side falls.

4 DISCUSSION

As most of the fall-related injuries occur when the

body hits the ground, the application of a pre-impact

fall detection approach along with fall impact

reduction systems for injury minimization will

provide useful intervention for elderly people

susceptible to faint falls (Wu, 2000; Davidson, 2004;

Lockhart, 2006; Ulert, 2002).

This study aimed to detect a fall before impact

using acceleration, angular velocity and angular

features. In this study, we achieve lead time of

approximately 530ms and 100% specificity.

Some previous studies showed 100% specificity.

However, they did not show 100% sensitivity (Wu,

2000; Bourke et al., 2008). In particular, sometimes

their algorithm made mistakes on jump or stand-sit

transition for fall. If using acceleration threshold

only, the jump might be mistaken for fall because

the variation of the acceleration was large. For the

stand-sit transition, especially during sitting to a

chair, the pattern of acceleration is similar to the

acceleration pattern of fall. Furthermore, there is

rapid variation of the acceleration pattern when hip

contacts to the chair. However, our algorithm using

the threshold of angle could avoid these wrong

recognitions.

In the assessment of successful balance recovery

from complete loss of balance in fall, Thelen et al.

Pre-impactFallDetectionusingWearableSensorUnit

209

Table 2: Peak acceleration, angular velocity and vertical angle during falls and ADLs.

Trials Acceleration (g)

Angular velocity (°/s) Angle (°)

Pitch Roll Sagittal Lateral

Falls

Backward

4.1 ± 0.6 300.3 ± 59.7 45.7 ± 14.2 94.2 ± 4.7 4.7 ± 3.2

Forward

4.5 ± 0.5 220.6 ± 41.6 75.9 ± 17.2 89.6 ± 9.2 8.4 ± 4.6

Side

4.4 ± 0.6 121.2 ± 13.7 419.4 ± 61.3 6.69 ± 2.7 75.95 ± 12.4

ADLs

Sit-Stand

1.4 ± 0.2 110 ± 23.1 8.9 ± 7.6 31.1 ± 6.1 2.6 ± 1.3

Stand-Sit

2.2 ± 0.3 392.3 ± 61.3 11.2 ± 6.1 11.63 ± 3.1 1.12 ± 2.7

Sit-Lying

1.1 ± 0.1 80.7 ± 31.7 15.3 ± 3.8 90.3 ± 8.3 4.3 ± 2.8

Walking

2.1 ± 0.2 50.1 ± 10.9 59.3 ± 14.9 1.4 ± 6.2 2.1 ± 3.1

Jump

7.5 ± 1.1 421.2 ± 149.1 102.3 ± 62.1 27.3 ± 2.1 3.6 ± 3.8

Running

4.2 ± 0.9 132.8 ± 45.7 98.2 ± 34.9 11.5 ± 9.7 2.6 ± 4.1

(1997) found that the maximum lean angle where

subjects could recover balance with a single forward

step averaged 32.5° for young men and 23.9° for

older men. Therefore, it can be noted that our

threshold 30° of sensor angle is very well within the

limits of balance recovery during the fall process.

For lying activities in ADLs, there was a small

variation in acceleration and no wrong recognition

was found even though the angle changed to 90°.

It should be pointed out that all activities tested

in this study was performed by healthy volunteers

aged below 30, since the experimental procedure

was not understandably suited for elderly subjects

who are at greater risk of suffering injury. The

movement of younger subjects is bound to differ

from that of the elderly population, who may have as

lower reaction time and lesser ability to rescue the

body from falling. In addition, our algorithm were

tested against a small range of fall types and ADLs.

Therefore, further tests are needed for other types of

falls such as tripping and slipping.

Nevertheless, our pre-impact fall detection

algorithm can be implemented in a wearable fall

injury minimization system to track a user’s body

movement and notify the fall impact reduction

device.

ACKNOWLEDGEMENTS

This study was supported by the National Research

Foundation of Korea (NRF) funded by the Ministry

of Education (2013H1B8A2032194).

REFERENCES

Annekenny, R. and O’Shea, D., 2002. Falls and syncope in

elderly patients. Clinics in Geriatric Medicine 18(2),

xiii–xiv.

Bourke, A. K., O’Donovan, K. J., O´ Laighin, G., 2008.

The identification of vertical velocity profiles using an

inertial sensor to investigate pre-impact detection of

falls. Medical Engineering & Physics 30(7), 937–946.

Davidson, M. E., 2004. System for protection from falls.

United States Patent Office Document,

US2004/0003455.

Gross, Y. T., Shimamoto, Y., Rose, C. L., Frank, B., 1990.

Why do they fall? Monitoring risk factors in nursing

homes. Journal of Gerontology and Nursing 16(6), 20–

25.

Hayes, W. C., Myers, E. R., Robinovitch, S. N., Van Den

Kroonenberg, A., Courtney, A. C., Kenny, R. A.,

O’Shea, D., 2002. Falls and syncope in elderly

patients. Clinics in Geriatric Medicine 18(2), xiii–xiv.

Lockhart, T. E., 2006. Fall-sensing systems, hip protector

systems, and other protective systems. United States

Patent Office Document, US2006/0049950A1.

Nevitt, M.C., Cummings, S. R., Hudes, E. S., 1991. Risk

factors for injurious falls: a prospective study. Journal

of Gerontology 46, M164–M170.

Nyan, M. N., Tay, F. E., Tan, A. W., Seah, K. H., 2006.

Distinguishing fall activities from normal activities by

angular rate characteristics and high-speed camera

characterization. Medical Engineering & Physics 28

(8), 842–849.

Peczalski, K., Wojciechowski, D., Sionek, P., Dunajski,

Z., Palko, T., 2006. Impedance cardiography vs clamp

pletysmography of Penaz in diagnostic of patients with

vasovagal syncope. Journal of Biomechanics 39(1),

S453.

Smeesters, C., Hayes, W. C., McMahon, T. A., 2001.

Disturbance type and gait speed affect fall direction

and impactlocation. Journal of Biomechanics 34(3),

309–317.

Thelen, D. G., Wojcik, L. A., Schultz, A. B., Ashton-

Miller, J. A., Alexander, N. B., 1997. Age differences

in using a rapid step to regain balance during a forward

all. Journals of Gerontology SeriesA: Biological

Sciences and Medical Sciences 52A(1), M8–M13.

Ulert, I. A., 2002. Hip protector. United States Patent

BIODEVICES2014-InternationalConferenceonBiomedicalElectronicsandDevices

210

Office Document, US2002/0078484.

Van den Kroonenberg, A. J., Hayes, W. C., McMahon, T.

A., 1996. Hip impact velocities and body

configurations for voluntary falls from standing height.

Journal of Biomechanics 29(6), 807–811.

Weatherall, M., 2004. Multifactorial risk assessment and

management programmes effectively prevent falls in

the elderly. Evidence-Based Healthcare and Public

Health 8(5), 270–272.

Wu, G., 2000. Distinguishing fall activities from normal

activities by velocity characteristics. Journal of

Biomechanics 33(11), 1497–1500.

Yamamoto, S., Tanaka, E., Ikeda, T., Kubouchi, Y.,

Harada, A., Okuizumi, H., 2006. Mechanical

simulation for hip fracture by a fall using multibody-

FE hybrid human model. Journal of Biomechanics

39(1), S89–S90.

Pre-impactFallDetectionusingWearableSensorUnit

211