Kinematic analysis of video captured falls experienced by older adults in long term care

Journal of Biomechanics 48 (2015) 911–920

Contents lists available at ScienceDirect

Journal of Biomechanics
journal homepage: www.elsevier.com/locate/jbiomech
www.JBiomech.com

Kinematic analysis of video-captured falls experienced by older adults
in long-term care
W.J. Choi a,n, J.M. Wakeling b, S.N. Robinovitch b,c
a

Department of Physical Therapy, Chapman University, 9401 Jeronimo Rd, Irvine, CA 92618, USA
Department of Biomedical Physiology and Kinesiology, Simon Fraser University, Burnaby, BC, Canada
c
School of Engineering Science, Simon Fraser University, Burnaby, BC, Canada
b

art ic l e i nf o

a b s t r a c t

Article history:
Accepted 15 February 2015

Falls cause 95% of hip and wrist fractures and 60% of head injuries in older adults. Risk for such injuries
depends in part on velocity at contact, and the time available during the fall to generate protective
responses. However, we have no information on the impact velocities and durations of falls in older
adults. We addressed this barrier through kinematic analysis of 25 real-life falls (experienced by 23
individuals of mean age 80 years (SD¼ 9.8)) captured on video in two long-term facilities.

All 25 falls involved impact to the pelvis, 12 involved head impact, and 21 involved hand impact. We
determined time-varying positions by digitizing each video, using direct linear transformations
calibrated for each fall, and impact velocities through differentiation.
The vertical impact velocity averaged 2.14 m/s (SD¼ 0.63) for the pelvis, 2.91 m/s (SD¼ 0.86) for the
head, and 2.87 m/s (SD¼ 1.60) for the hand. These values are 38%, 28%, and 4% lower, respectively, than
predictions from an inverted pendulum model. Furthermore, the average pelvis impact velocity was 16%
lower than values reported previously for young individuals in laboratory falling experiments. The
average fall duration was 1271 ms (SD¼ 648) from the initiation of imbalance to pelvis impact, and
583 ms (SD¼255) from the start of descent to pelvis impact.
These first measures of the kinematics of falls in older adults can inform the design and testing of fall
injury prevention interventions (e.g., hip protectors, helmets, and flooring).
& 2015 Elsevier Ltd. All rights reserved.

Keywords:
Falls
Biomechanics
Older adults
Hip fracture
Impact velocity
Injury
Kinematic analysis
Video

1. Introduction
Falls are the number one cause of injuries in older adults,
including at least 90% of hip fractures and wrist fractures, and 60%
of head injuries (Grisso et al., 1990; Harvey and Close, 2012; Palvanen
et al., 2000). The risk for injury during a fall depends in part on the
velocity at contact (or “impact velocity”) of the impacting body parts
(Majumder et al., 2008; Robinovitch et al., 1991). Accordingly, impact

velocity is a key input parameter for biomechanical testing of fall
injury prevention technology (e.g., hip protectors (Mills, 1996; Minns
et al., 2004a; Robinovitch et al., 2009), helmets (ASTM, 2007), and
compliant flooring (Knoefel et al., 2013; Laing and Robinovitch, 2009;
Minns et al., 2004b)). Risk for injury during a fall may also depend on
the time duration of the fall, which governs the faller's ability to
initiate and execute protective responses, such as arresting the fall
with the upper limbs (DeGoede et al., 2001; Robinovitch et al., 2005).

n

Corresponding author.
E-mail address: (W.J. Choi).

/>0021-9290/& 2015 Elsevier Ltd. All rights reserved.

Our current knowledge of the impact velocities and durations
associated with falls is limited to the results of laboratory studies
with young adults falling (from standing height) onto gym
mats (Feldman and Robinovitch, 2007; Hsiao and Robinovitch,
1998; Robinovitch et al., 2003). However, the kinematic patterns
observed in lab-based falling experiments with young adults may
differ substantially from real-life falls in older adults, due to
differences in the situational and environmental context of falls,
or age-related changes in physiological factors such as physical
and cognitive function, medication use, and disease. On the one
hand, real-life falls in older adults may generate impact velocities
that are higher than those observed in laboratory-based falls in
healthy young adults, due to age-related declines in sensorimotor
and cognitive function, and a corresponding absence or declines in

balance recovery responses (e.g., stepping or grasping) and fall
protective responses (e.g., upper limb fall arrest). On the other
hand, the perturbation conditions associated with real-life falls,
which tend to be caused by internal versus external perturbations
(Robinovitch et al., 2013), may cause them to less severe than
laboratory-based falls which have used large external perturbations to overcome participants' ability to recover balance.


912

W.J. Choi et al. / Journal of Biomechanics 48 (2015) 911–920

Our purpose in the current study was to document the impact
velocities of key body sites (hip, head and hand) during real-life falls
in older adults. We collected and conducted kinematic analysis of
video footage of real-life falls experienced by older adults in longterm care (LTC) facilities. We then determined the impact velocities
of the hip, head and hand, and the time duration of the falls. We
compare our results to previous studies with young adults, and to
theoretical predictions from simple mathematical models.

2.2. Fall duration
The questionnaire required the team to estimate the exact video frames
corresponding to the onset of imbalance leading to the fall, the initiation of the
descent stage of the fall (defined as one video frame after the foot contacted the
ground in the last recovery step, if any), and the first occurrence of impact to the hand
(s), head and pelvis. We report two estimates of fall duration before impact to the
body parts: the “total fall duration”, defined by the interval between the onset of
imbalance and initial impact to the body part, and the “descent duration,” defined as
the interval between the onset of fall initiation and impact.

2. Methods

2.3. Impact velocities

2.1. Real-life fall library

To estimate impact velocities, we manually digitized landmark of the pelvis
(anterior superior iliac spine), head (ear or forehead) and hand (palm), using a Matlab
routine developed by Hedrick (2008). We digitized each frame over the interval
starting one frame before fall initiation and ending one frame after impact of the
corresponding body part (Fig. 1(a)). We then applied a two-dimensional direct linear
transform (2D DLT) to reconstruct those points as position coordinates in the object
space (Hedrick, 2008; Meershoek, 1977). Finally, we used finite difference to estimate
time-varying vertical and horizontal velocities. The resulting velocity—time traces
were fit with a fifth-order polynomial (Fig. 2) using Matlab's polyfit function, the
approach used in falling experiments by van den Kroonenberg et al. (1996) to fit
vertical displacement versus time curves for the hip during falls from standing. The
vertical impact velocity was estimated as the maximum value of the curve fit, based on
previous observations (and theoretical considerations) that the peak downward
velocity occurs very near to the instant of contact (Feldman and Robinovitch, 2007;
Hsiao and Robinovitch, 1998). We also report values of the peak horizontal velocity
and the magnitude of horizontal velocity at the instant of peak vertical velocity.
While our 2D DLT procedure corrected for lens distortion, an important
limitation of the technique is the potential for “perspective errors” arising when
the digitized points of interest move outside the calibrated image plane.
In an attempt to minimize these errors, we determined DLT calibration
coefficients specific to each fall video, by visiting the site of each fall, and recording
images of a flat calibration panel from the surveillance camera that captured the fall

This study builds on recent descriptive reports by our team on the circumstances of

falls captured on video cameras in two long-term care (LTC) facilities (Robinovitch et al.,
2013; Yang et al., 2013). Between April 2007 and February 2013, we partnered with two
LTC facilities in the Vancouver area (Delta View, a 312-bed facility in Delta, BC, and New
Vista, a 236-bed facility in Burnaby, BC) to capture video footage of real-life falls in older
adults (Robinovitch et al., 2013). Delta View had a network of 216 digital cameras, while
New Vista had 48. Cameras were located in common areas (dining rooms, hallways, and
lounges) and not bedrooms or bathrooms. Each video was recorded at 30 frames
per second and a resolution of 640 by 480 pixels or 720 by 480 pixels (Fig. 1). The study
was approved by the Office of Research Ethics at SFU. Each resident provided written
consent to the facilities for video capture of their images, and these data were shared as
secondary data with the research team. Additional written consent was secured from
individuals to share their images.
Each fall video was initially analyzed by a team of three experts (research
assistants and graduate students trained by co-author SNR) using a structured,
validated questionnaire (Yang et al., 2013). The questionnaire probed the cause of fall,
the activity at the time of the fall, the initial direction of the fall (forward, backward,
sideways, or straight down), the landing configuration (forward, backward, or
sideways), the occurrence (if any) of stepping responses, and the occurrence of
impact to the hand(s), knee(s), head and pelvis.

Fig. 1. Video snapshots showing: (a) forward fall by older adult in long-term care (LTC); (b) forward fall by young adult in the laboratory environment; (c) backward fall by
older adult in LTC; and (d) backward fall in young adult in the lab. The far-right panel illustrates the 25-marker calibration panel, placed at the exact location of the fall (in the
lab or LTC facility), and oriented in the plane of the fall. The white dots in Fig. 1(a) indicate digitized pelvis (ASIS) landmarks.


W.J. Choi et al. / Journal of Biomechanics 48 (2015) 911–920

913

Fig. 2. Traces from typical trials of the velocity of the pelvis versus time for: (a) forward fall by an older adult in long-term care (LTC); (b) backward fall by an older adult in

LTC; (c) forward fall by a young adult in the lab; and (d) backward fall by a young adult in the lab. Vertical velocities are shown in dashed lines/ circles, and horizontal
velocities in solid lines/squares.
(see far-right images in Fig. 1). The panel had dimensions 160 cm  160 cm, and
contained a 5 Â 5 grid of 10 cm diameter circular markers spaced 40 cm apart. The
panel was placed with the bottom surface flush to the ground, centered at the
midpoint of the faller's feet (at the moment of fall initiation), and oriented in the plane
of the fall (defined by the line connecting the mid-point of the feet and the location of
the head at the moment of pelvis impact).
2.4. Laboratory measures of accuracy
We tested the accuracy of our velocity estimates through laboratory falls with
an inverted pendulum and a human participant. Each trial was captured with an
8-camera motion capture system recording at 250 Hz (Motion Analysis Corp., Santa
Rosa, CA, USA), and a single surveillance camera identical to the type used in our
partnering LTC sites recording at 30 Hz (model WVC210, Cisco Systems, San Jose, CA).
The surveillance camera was placed at a height of 2.6 m and horizontal distance of 5 m
from the site of the fall, which was typical of the falls we captured in the two long term
care facilities (although there was variability in this distance for the real-life falls).
The pendulum consisted of a 1.57 m long aluminum rod of uniform mass
distribution, connected to a low-friction hinge joint at the floor. Reflective markers
were placed on the midpoint (representing the pelvis) and top end (head) of the
rod. The pendulum was released from vertical and descended over a 901 arc before
impacting the ground.
Human falls were conducted for three falling directions: forward, backward,
and sideways (Fig. 1). In all trials, participants self-initiated the fall, and were
instructed to fall naturally. Reflective markers were placed on the anterior superior
iliac spines, greater trochanters, sacrum, wrists, elbows, shoulders and forehead.
Trials were conducted at five falling angles with respect to the axis of the
surveillance camera: 301 (nearly toward the camera), 601, 901 (perpendicular to the
camera axis, providing a sideways view), 1201, and 1501 (nearly away from the camera).
A single trial was acquired for each camera angle, in (for human falls) each of the three

fall directions.

standard deviation in the offset error across the five camera angles. We regarded
the technique as acceptable for a given fall direction if the standard deviation in the
offset error was 0.35 m/s or less, which would reflect a 95% confidence interval less
than or equal to 7 0.7 m/s in the estimated impact velocity. We regarded the mean
offset error (for a given direction) as a fixed bias, and subtracted direction-specific
mean errors from peak velocities estimated from video analysis.
We also excluded falls with movements that were clearly different than those
involved in our calibration trials. These included falls with significant rotation during
descent, falls directly toward or away from the camera, falls directed straight down,
and falls involving impact to objects other than the floor (e.g., walls or furniture). We
also excluded falls not involving pelvis impact, or where the pelvis was occluded
from camera view during descent or impact, since our primary objective at the onset of
the study was characterizing the severity of impact to the pelvis during falls.

2.6. Analysis of video-captured falls in older adults
We provide descriptive results (means and standard deviations) for the vertical and
horizontal impact velocity of the pelvis, head and hand. We focus our attention more on
vertical than horizontal velocity, as the stronger indicator of risk for serious injury
(van den Kroonenberg et al., 1996). We also report total fall durations and descent
durations for each body part. We compare measured vertical velocities
toffi theoretical
pffiffiffiffiffiffiffiffi
estimates based on free fall of a falling mass (where impact velocity¼ 2gh, where g is
2
gravitational acceleration of 9.81 m/s , and h is the vertical descent distance of the body
part), andpan
inverted
pendulum with uniformly-distributed mass (where impact

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
velocity¼ ð3=2Þgh). We also conducted regression analysis to test whether impact velocity
is associated with fall height.

3. Results

2.5. Inclusion criteria for video-captured falls in older adults

3.1. Laboratory falls

We used the results from our laboratory-based human falls to guide the
selection of video-captured falls in older adults for analysis, based on fall direction.
In particular, for each trial, we calculated the offset error, defined as the difference
between the impact velocity estimate from 3D motion capture and the 2D single
video camera images. We then determined, for a given fall direction, the mean and

In our laboratory experiments with the inverted pendulum (Fig. 3),
the average difference (across the five falling angles) between the 2D
DLT and 3D motion capture techniques was 0.008 m/s (SD¼0.04)
for the vertical impact velocity, and À 0.11 m/s (SD¼0.07) for the


914

W.J. Choi et al. / Journal of Biomechanics 48 (2015) 911–920

Fig. 3. Agreement from laboratory experiments between the 2D DLT technique and 3D motion capture for vertical and horizontal impact velocities. Results are shown for the
mid-point of a 1.57 m length pendulum, and for the pelvis for a human participant falling in the forward, backward, and sideways directions (at five different camera angles;
see text for explanation). Sideways falls in human exhibited larger variability between camera angles (vertical velocity SD ¼ 0.59 m/s, horizontal velocity SD ¼0.54 m/s) than

forward and backward falls. The individual plots clearly show that 2 SD is less than 0.7 m/s in all but the sideways direction.


W.J. Choi et al. / Journal of Biomechanics 48 (2015) 911–920

horizontal impact velocity (where a negative mean reflects overestimation from the 2D DLT technique).
In laboratory-based human falls (Fig. 3), the DLT technique met
our criteria for acceptable accuracy for forward and backward falls,
but not sideways falls. The mean difference in vertical impact
velocity between the 2D DLT and 3D motion capture techniques
was À 0.26 m/s (SD¼ 0.21) for forward falls, À0.06 m/s (SD¼0.21)
for backward falls, and À 1.01 m/s (SD was 0.59) for sideways falls.
The mean difference in horizontal impact velocity was À 0.15 m/s
(SD¼0.32) for forward falls, À 0.16 m/s (SD¼ 0.18) for backward falls,
and À 0.13 m/s (SD¼0.54) m/s for sideways falls. For both human
and inverted pendulum falls, the accuracy was not influenced by the
falling angle with respect to the camera axis (p40.5 by linear
regression). For human falls, the mean error was not different
(p40.1) between forward and backward falls, and was equal to
À0.16 m/s for both vertical and horizontal velocity. The pooled SD
for forward and backward falls was 0.23 m/s for vertical velocity and
0.24 m/s for horizontal velocity.
3.2. Falls by older adults
Between April 2007 and February 2013, we captured 813 falls
experienced by 306 individuals (Fig. 4). Based on our acceptance
criteria and the results from our laboratory-based falling experiments, we excluded cases where the initial fall direction was
sideways (n¼ 152). We excluded an additional 636 cases based on
other exclusion criteria, as described in Fig. 4.
Our final analysis included 25 falls (23 from Delta View, 2 from
New Vista) experienced by 23 older adults (Tables 1 and 2). There

were 21 backward falls and 4 forward falls, all of which involved
pelvis impact. The average age of the faller was 80.3 yrs (SD¼9.8),
and 61% (n¼ 14) were female. The most common cause of imbalance

915

was incorrect weight shifting (12 of 25 cases), followed by hit/bump
(7 cases). Trips, collapses, and loss-of-support each accounted for
2 falls. The most common activity at the time of falling was standing
(14 of 25 cases), followed by walking (8 cases), and transferring from
standing to sitting (3 cases). Two falls occurred while using a walker
(video IDs 23 and 24). Stepping after the onset of imbalance occurred
in 16 of 25 falls.
Head impact occurred in 48% of cases (n ¼12; all 4 forward falls,
and 8 of 21 backward falls), and hand impact occurred in 84%
(n ¼21; all 4 forward falls, and 17 of 21 backward falls). In 76% of
cases involving hand impact (n ¼ 16), the hand impacted before the
pelvis or head. All four forward falls involved impact to the knee
(s) before the pelvis.
Table 2 reports estimated impact velocities for each fall, after
subtracting direction-specific mean offset errors (for vertical velocity:
0.26 m/s for forward falls and 0.06 m/s backward falls; for horizontal
velocity: 0.15 m/s for forward falls and 0.16 m/s backward falls).
Over the 25 falls by older adults, the vertical impact velocity
averaged 2.14 m/s (SD¼ 0.63) for the pelvis, 2.91 m/s (SD¼ 0.86)
for the head, and 2.87 m/s (SD¼1.60) for the hand (Table 2 and
Figs. 5 and 6). For eight backward falls involving impact to both the
pelvis and head, the vertical impact velocity was 2.67 (SD¼0.82) for
the head and 1.98 (SD¼0.45) m/s for the pelvis. The horizontal
impact velocity averaged 1.16 m/s (SD¼1.42) for the pelvis, 2.64 m/s

(SD¼1.12) for the head, and 1.52 m/s (SD¼1.14) for the hand. The
total fall duration averaged 1271 ms (SD¼648) for the pelvis, 1730 ms
(SD¼805) for the head, and 1188 ms (SD¼702) for the hand. The
descent duration averaged 593 ms (SD¼255) for the pelvis, 757 ms
(SD¼217) for the head, and 479 ms (SD¼230) for the hand.
The vertical impact velocity of the pelvis averaged 2.19 m/s
(SD ¼0.61) in trials where the hand(s) impacted before the pelvis,
compared to 2.41 m/s (SD ¼ 0.85) in falls not involving hand

Fig. 4. Fall video selection process. Among 813 fall videos captured, 25 forward and backward falls were selected for analysis.


916

Table 1
Participant characteristics and descriptive data for 25 falls from 23 older adults.
Faller
ID

Age Sex Body mass
(kg)

1
2
3
4
5
6
7
8

9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

1
2
3
4
5
6
7
8
9
10
11
2

12
13
14
15
16
17
18
18
19
20
21
22
23

84
93
71
70
69
64
90
82
84
63
84
93
87
72
69
88

86
74
75
75
86
100
79
91
79

F
F
M
M
F
M
M
F
M
M
M
F
F
F
F
F
M
F
M
M

M
F
F
F
F

Parkinson's
disease

Alzheimer's
disease

Stroke Hypertension COPD Diabetes Fall
direction

√
√
√
√
√

57.6
58.1
72.5






√

√
√

61.1




√
√

85.2
58.1
68.4
50.9
79.1
65.7
88.9
68






45.4





√

√
√

√
√
√
√

√
√
√
√

√
√
√
√

√
√

√
√

√
√

√
√

√

√
√

F
F
F
F
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B

B
B
B

Pelvis
impact

Head
impact

Hand
impact

Knee
impact

Cause of
fall

Activity
at the
time of
fall

Stepping
response

√
√
√

√
√
√
√
√
√
√
√
√
√
√
√
√
√
√
√
√
√
√
√
√
√

√
√
√
√

√
√

√
√
√
√
√

√
√
√
√

T
T
B
IT
B
B
LOS
B
IT
IT
IT
IT
C
IT
B
LOS
IT
IT
C

IT
IT
IT
IT
B
B

W
W
W
W
S
S
S
S
S
T
T
S
S
W
S
T
W
W
S
S
W
S
S

S
S

No
Yes
Yes
No
No
Yes
No
Yes
Yes
No
No
Yes
Yes
Yes
No
Yes
Yes
Yes
No
Yes
No
Yes
Yes
Yes
Yes

√

√

√
√
√
√

√
√

√
√
√
√
√
√
√
√
√
√
√
√
√
√

Fall direction: F ¼forward, B ¼backward; Cause of fall: T ¼ trip/stumble, B¼ hit/bump, IT ¼incorrect transfer, LOS¼loss of support, C¼ leg collapse; Activity at the time of fall: W ¼ walking, S¼standing, T¼ transferring;
data; COPD ¼ chronic obstructive pulmonary disease.

Injury

noted

Head
Head

Head
Head
Head

 ¼missing

W.J. Choi et al. / Journal of Biomechanics 48 (2015) 911–920

Video
ID


Table 2
Impact velocities, fall durations, and fall heights for 25 falls from 23 older adults.
Video ID

Faller
ID

Head impact

Peak
horizontal
velocity
(m/s)

Total fall
Horizontal
duration
velocity at
peak vertical (ms)
velocity (m/s)

Descent
duration
(ms)

Fall
height
(cm)

1
2
3
4

1.68
1.29
1.70
1.98

7.50
1.68
4.32
1.39

2.34
1.42
3.63
1.39

1370
1290
3067
1097

1370
645
533
1097

125.3
99.6
72.7
99.8

5
6
7
8
9
10
11
2
12

13
14
15
16
17
18
18
19
20
21
22
23

1.18
1.74
3.51
3.37
2.13
2.46
2.02
3.03
1.68
1.71
1.75
1.29
2.62
1.80
3.19
2.06
2.05

2.50
1.89
2.29
2.58
2.14
0.63

0.49
1.99
7.23
2.18
0.63
0.02
1.10
0.77
0.65
1.10
0.16
0.84
1.31
0.95
0.17
0.09
1.60
0.64
0.82
1.15
0.96
1.59
1.95

0.38
0.07
6.72
2.12
0.61
0.09
1.09
0.40
0.49
1.10
0.14
0.79
0.86
0.95
0.07
0.08
1.13
0.22
0.82
1.07
0.96
1.16
1.42

500
1125
848
1074
3063

1000
467
1233
1067
1333
600
621
1677
1400
933
1167
800
1833
1433
1033
1733
1271
648

500
625
848
815
500
571
467
400
167
367
600

621
323
533
367
767
800
400
467
567
467
593
255

55.1
78.7
103.6
106.2
72.3
94.4
91.0
83.3
57.2
63.9
84.2
66.0
79.3
85.3
77.7
74.3
88.7

62.0
67.4
84.7
65.2
81.5
17.0

Peak
vertical
velocity
(m/s)

Peak
horizontal
velocity
(m/s)

Hand impact
Total fall
Horizontal
duration
velocity at
peak vertical (ms)
velocity (m/s)

Descent
duration
(ms)

Fall

height
(cm)

Peak
vertical
velocity
(m/s)

Peak
horizontal
velocity
(m/s)

Total fall
Horizontal
duration
velocity at
peak vertical (ms)
velocity (m/s)

Fall
height
(cm)

a

a

a

a

a

a

a

a

a

a

a

3.52
2.80
4.25

2.68
4.36
1.87

2.68
3.86
1.73

1194
3067

1129

548
533
1129

84.9
100.7
164.0

2.47
3.00
2.84

3.95
4.42
2.20

3.95
2.89
2.18

774
2833
968

129
300
968

28.1
67.6
101.0

3.29

3.49

2.57

1125

625

96.4

7.38
1.78
4.25

2.24
1.19
3.20

2.24
0.77
0.57

406
1000

636

406
500
636

34.5
65.7
134.0

2.12

3.14

2.89

3313

750

86.0

1.50

1.35

1.32

3063

500

66.7

2.67
2.00
2.30
1.85
4.47

1.83
1.77
1.29
1.30
1.37

1.64
1.22
0.60
1.19
0.52

400
1167
1033
1267
667

400
333

133
300
667

51.8
59.9
54.4
45.6
107.7

2.12
2.07
4.31
3.04

1.25
2.32
4.20
4.28

0.31
2.11
3.80
4.11

1133
862
1871
1633

1133
862
516
767

109.4
95.7
139.1
129.1

1.96

2.49

1.84

1767

800

110.7

2.48
2.91
0.86

3.12
3.02
1.02

3.10
2.64
1.12

1933
1730
805

667
757
217

90.5
109.7
24.9

a

a

a

a

a

a

2.28
5.56

1.03
3.83

3.22
0.33
0.10
3.50

3.22
0.33
0.48
3.50

1333
900
1267
833

467
333
867
833

89.3
80.3
61.5
65.4

1.69
1.81

1.76
2.87
1.60

0.76
2.03
1.25
1.96
1.20

0.67
0.75
0.88
1.52
1.14

1433
967
1633
1188
702

467
500
367
479
230

57.8
40.6

55.1
66.7
26.2

Total fall duration ¼time interval between the moment of imbalance and impact; Descent duration ¼time interval between fall initiation and impact; Fall height¼ vertical descent distance from fall initiation to impact;
a

Descent
duration
(ms)

 ¼data not available.

W.J. Choi et al. / Journal of Biomechanics 48 (2015) 911–920

Peak
vertical
velocity
(m/s)
Forward falls
1
2
3
4
Backward falls
5
6
7
8
9

10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Mean
Standard deviation

Pelvis impact

Not able to digitize.

917


918

W.J. Choi et al. / Journal of Biomechanics 48 (2015) 911–920

Fig. 5. Mean values from the 25 falls in older adults (with standard deviations shown as error bars) of: (a) vertical impact velocity; (b) horizontal impact velocity; (c) total fall
duration; and (d) descent duration. In each case, values of shown for each of the pelvis, head and hand.

impact, and 2.19 m/s (SD ¼0.58) in trials where steps occurred
after imbalance versus 2.31 m/s (SD ¼0.81) in falls not involving
steps (Table 2).
When compared to theoretical predictions based on free-fall
from each measured fall height (Table 2 and Fig. 7), our vertical
pelvis impact velocities averaged 46.0% (SD¼14.95) lower than
predictions from a falling mass model, and 38.0% (SD¼17.3) lower
than predictions from an inverted pendulum model. Similarly, our
vertical head and hand impact velocities averaged 37.4% (SD¼15.3)
and16.9% (SD¼ 55.4) lower than free-fall predictions, and 27.7%
(SD¼17.7) and 4.0% (SD¼ 63.9) lower than pendulum fall predictions, respectively. Furthermore, from regression analysis (SPSS,
Version 18.0), we found that fall height associated with the vertical
impact velocity of the head (v¼0.022nhþ0.5, R2 ¼0.403, p¼ 0.036),
but not of the pelvis (p¼0.19) or hand (p¼0.41).

4. Discussion
This is the first study to our knowledge to report impact
velocities and fall duration from real-life falls in older adults. Our
results provide important baseline measures of fall severity for the
design and assessment (through mechanical testing systems or
mathematical models) of interventions for fall injury prevention,
including wearable hip protectors, helmets and compliant flooring.
Impact velocity is a measure of fall severity that is important for
the design and testing of injury prevention strategies. Our measured
vertical impact velocities for older adults averaged 16% lower than
the mean value for the pelvis (of 2.55 m/s (SD¼ 0.85)) and 9% greater
than the mean value for the wrist (2.64 m/s (SD¼0.66)) reported by

Hsiao and Robinovitch (1998) from a laboratory study with young
adults falling on gym mats after receiving a sudden (slip) perturbation. This previous study included 20 backward and 11 sideways falls
in the analysis, and only reported average values for all trials, without

separating the results by fall direction (Hsiao and Robinovitch, 1998).
Furthermore, our vertical impact velocities averaged 38% lower for
the pelvis, and 28% lower for the head, than theoretical predictions
from an inverted pendulum model, based on fall height. Moreover,
the fall height associated with vertical impact velocity of the head,
but not of the pelvis or hand.
These results suggest that, in the falls we analyzed, older adults
utilized mechanisms to absorb energy during descent, and reduce
their impact velocity (and risk for injury). These included attempts to
recover balance by stepping (which occurred in 64% of falls), and
impacting the ground with the hands before the pelvis or head
(which occurred in 84% of falls). In previous falling experiments with
young adults, pelvis impact velocities were decreased 22% by taking a
step after imbalance, and 18% by impacting the hands before the
pelvis (Feldman and Robinovitch, 2007). While our small sample
precluded meaningful statistical analysis, our trends agree with these
findings. Pelvis impact velocities averaged 9% lower in falls involving
hand impact (compared to no hand impact), and 5% lower in falls
involving stepping (compared to no stepping). Additional mechanisms beyond the scope of this study may have contributed to velocity
reduction, including squatting during descent and contacting the
pelvis with the trunk relatively upright (Robinovitch et al., 2004).
The fall duration is of interest since it reflects the time available for
the faller to initiate and execute protective responses to avoid injury
during landing. Our study shows that a duration of nearly 1200 ms is
available to initiate protective responses between the moment of
imbalance, and the instant of impact to the pelvis (which occurred on

average at 1271 ms) or hand (which occurred on average at 1188 ms).
This is considerably longer than the fall durations reported by Hsiao
and Robinovitch (1998) in their laboratory falls due to sudden slip
perturbations, where the average interval between the onset of the
perturbation and pelvic contact was 715 ms (SD¼160), while wrist
contact was 680 ms (SD¼116). This reflects that real-life falls in older
adults occur over a considerably slower time interval than falls


W.J. Choi et al. / Journal of Biomechanics 48 (2015) 911–920

919

Fig. 6. Sample traces of the vertical and horizontal velocity of the pelvis, head, and hand versus time for (a) forward fall and (b) backward fall by older adults in long-term
care. Both falls resulted in impact to the hand, pelvis, and head. In the forward fall, impact occurred first to the hand. This seemed to reduce the subsequent vertical velocity
of the pelvis and head, which impacted the ground near-simultaneously. During the backward fall, impact first occurred near-simultaneously to the pelvis and hand. This
seemed to reduce the vertical velocity of the head, before a final rapid increase as the trunk rotated downward.

recorded in laboratory experiments with young adults, where a rather
severe, sudden perturbation is necessary to overcome balance recovery responses. Another study (Robinovitch et al., 2005) found that the
time required for older adults to move their hands into a protective
position to arrest a fall averages 615 ms (SD¼ 88). This is well below
our average value of total fall duration but similar to our mean
descent duration. In our study, upper limb protective responses were
typically initiated soon after the onset of imbalance (Fig. 6), which
likely contributed to the observation of hand impact in 84% of falls.
There are important limitations to our study. Our results are based
on falls experienced by residents in LTC, and may not apply to
healthier community dwelling older adults, or young adults. Our small
sample size prevented us from examining how falls associate with

intrinsic factors such as physical and cognitive function or medications.
Larger studies are required to relate the kinematics of falls to the
clinical context. We only included falls that involved pelvis impact,
leading us to exclude many forward falls. Our video footage was
collected at 30 fps, and therefore our resolution in detecting fall
initiation and impact times was limited to the duration of one frame
of the video (33 ms), or about one-fifteenth (7%) of the shortest
descent duration we report (479 ms for the hand). We calculated
velocities after fitting displacement versus time traces with a fifthorder polynomial, which may have resulted in filtering or loss of valid
kinematic information. However, our approach was similar to that
used in previous video-based laboratory measures of fall impact
velocities in humans (van den Kroonenberg et al., 1996). Furthermore,
we found that a fifth-order polynomial provided the best agreement to

velocity estimates from 3D motion capture (recording at 250 Hz) in
our inverted pendulum calibration tests. Furthermore, analyzing the
complex movements of falls from planar video is challenging, due to
the out-of-plane motions of the body segments that often accompany
during descent. In our laboratory falls, we found acceptable accuracy
in our velocity estimates for forward and backward falls, where the
trajectory of body parts tended to remain parallel to the calibration
plane. However, measurement accuracy was unacceptable for sideways falls (where knee and trunk flexion often caused out-of-plane
movement of the pelvis), which were excluded from analysis. The
exclusion of sideways falls was unfortunate given that hip fractures are
most likely to occur from sideways falls (Greenspan et al., 1994; Nevitt
and Cummings, 1993). Vertical impact velocities for the pelvis averaged 2.23 for backward falls and 1.66 m/s for forward falls, and one
might expect similar values for sideways falls. Future studies might
test this hypothesis by capturing sideways falls with 3D cameras, or
with multiple cameras and 3D analysis techniques.
In summary, based on analysis of 25 video-captured falls experienced by 23 older adults in long-term care, we found that the vertical

impact velocity averaged 2.14 m/s for the pelvis, and 2.91 m/s for the
head. These values are 38% and 28% lower, respectively, than theoretical predictions from an inverted pendulum model based on fall
height. Furthermore, the average vertical impact velocity of the pelvis
was 16% lower than values reported for young individuals in laboratory falling experiments. The duration of the fall averaged 1271 ms
from the moment of imbalance, and 583 ms from the start of descent,
to the instant of pelvis impact. These first measures of the kinematic


920

W.J. Choi et al. / Journal of Biomechanics 48 (2015) 911–920

influence this work, including employment, consultancies, stock
ownership, honoraria, paid expert testimony, patent applications/
registrations, and grants or other funding.

Acknowledgments
This research was supported by the operating grants from the
Canadian Institutes for Health Research (funding reference nos.
AMG-100487 and TIR-103945).
References

Fig. 7. Comparison for the 25 falls by older adults between measured and model
predictions of vertical impact velocities (based on fall height) for (a) pelvis,
(b) head, and (c) hand. On average, the measured vertical impact velocities for
the pelvis were 48.0% (SD ¼ 14.2) lower than free-fall model predictions and 40.0%
(SD ¼16.4) lower than inverted pendulum predictions. The vertical impact velocities for the head were 38.4% (SD¼ 16.5) and 28.8% (SD¼ 19.0) lower than free-fall
and inverted pendulum model predictions, respectively, and the vertical hand
impact velocities of the hand were 18.8% (SD ¼ 55.3) and 6.3% (SD¼ 63.9) lower
than free-fall and inverted pendulum model predictions, respectively.

profiles of real-life falls in older adults should inform the development
and testing of fall prevention technology, including wearable hip
protectors, helmets, and compliant flooring, and contribute to the
design of exercise programs to train fall protective responses.
Conflict of interest statement
None of the authors above have any financial or personal relationships with other people or organizations that could inappropriately

ASTM. A.S.f.T.a.M., 2007. Standard Performance Specification for Ice Hockey
Helmets.
DeGoede, K.M., Ashton-Miller, J.A., Liao, J.M., Alexander, N.B., 2001. How quickly
can healthy adults move their hands to intercept an approaching object? Age
and gender effects. J. Gerontol. Ser. A Biol. Sci. Med. Sci. 56, 584–588.
Feldman, F., Robinovitch, S.N., 2007. Reducing hip fracture risk during sideways
falls: evidence in young adults of the protective effects of impact to the hands
and stepping. J Biomech. 40, 2612–2618.
Greenspan, S.L., Myers, E.R., Maitland, L.A., Resnick, N.M., Hayes, W.C., 1994. Fall
severity and bone mineral density as risk factors for hip fracture in ambulatory
elderly. J. Am. Med. Assoc. 271, 128–133.
Grisso, J.A., Schwarz, D.F., Wishner, A.R., Weene, B., Holmes, J.H., Sutton, R.L., 1990.
Injuries in an elderly inner-city population. J. Am. Geriatr. Soc. 38, 1326–1331.
Harvey, L.A., Close, J.C., 2012. Traumatic brain injury in older adults: characteristics,
causes and consequences. Injury 43, 1821–1826.
Hedrick, T.L., 2008. Software techniques for two- and three-dimensional kinematic
measurements of biological and biomimetic systems. Bioinspir. Biomim. 3,
034001.
Hsiao, E.T., Robinovitch, S.N., 1998. Common protective movements govern unexpected falls from standing height. J. Biomech. 31, 1–9.
Knoefel, F., Patrick, L., Taylor, J., Goubran, R., 2013. Dual-stiffness flooring: can it
reduce fracture rates associated with falls? J. Am. Med. Dir. Assoc. 14, 303–305.
Laing, A.C., Robinovitch, S.N., 2009. Low stiffness floors can attenuate fall-related

femoral impact forces by up to 50% without substantially impairing balance in
older women. Accid. Anal. Prev. 41, 642–650.
Majumder, S., Roychowdhury, A., Pal, S., 2008. Effects of trochanteric soft tissue
thickness and hip impact velocity on hip fracture in sideways fall through 3D
finite element simulations. J. Biomech. 41, 2834–2842.
Meershoek, L., 1977. Matlab Routines for 2D Analysis with Non-Perpendicular
Camera Angle. International Society of Biomechanics.
Mills, N.J., 1996. The biomechanics of hip protectors. Proc. Inst. Mech. Eng. Part H J.
Eng. Med. 210, 259–266.
Minns, J., Dodd, C., Gardner, R., Bamford, J., Nabhani, F., 2004a. Assessing the safety
and effectiveness of hip protectors. Nurs. Stand. 18, 33–38.
Minns, J., Nabhani, F., Bamford, J.S., 2004b. Can flooring and underlay materials
reduce hip fractures in older people? Nurs. Older People 16, 16–20.
Nevitt, M.C., Cummings, S.R., 1993. Type of fall and risk of hip and wrist fractures:
the study of osteoporotic fractures. The Study of Osteoporotic Fractures
Research Group. J. Am. Geriatr. Soc. 41, 1226–1234.
Palvanen, M., Kannus, P., Parkkari, J., Pitkajarvi, T., Pasanen, M., Vuori, I.,
Jarvinen, M., 2000. The injury mechanisms of osteoporotic upper extremity
fractures among older adults: a controlled study of 287 consecutive patients
and their 108 controls. Osteoporos. Int. 11, 822–831.
Robinovitch, S.N., Brumer, R., Maurer, J., 2004. Effect of the “squat protective
response” on impact velocity during backward falls. J. Biomech. 37, 1329–1337.
Robinovitch, S.N., Evans, S.L., Minns, J., Laing, A.C., Kannus, P., Cripton, P.A., Derler, S.,
Birge, S.J., Plant, D., Cameron, I.D., Kiel, D.P., Howland, J., Khan, K., Lauritzen, J.B.,
2009. Hip protectors: recommendations for biomechanical testing-an international consensus statement (Part I). Osteoporos. Int. 20, 1977–1988.
Robinovitch, S.N., Feldman, F., Yang, Y., Schonnop, R., Leung, P.M., Sarraf, T.,
Sims-Gould, J., Loughin, M., 2013. Video capture of the circumstances of falls
in elderly people residing in long-term care: an observational study. Lancet
381, 47–54.
Robinovitch, S.N., Hayes, W.C., McMahon, T.A., 1991. Prediction of femoral impact

forces in falls on the hip. J. Biomech. Eng. 113, 366–374.
Robinovitch, S.N., Inkster, L., Maurer, J., Warnick, B., 2003. Strategies for avoiding
hip impact during sideways falls. J. Bone Miner. Res. 18, 1267–1273.
Robinovitch, S.N., Normandin, S.C., Stotz, P., Maurer, J.D., 2005. Time requirement
for young and elderly women to move into a position for breaking a fall with
outstretched hands. J. Gerontol. Ser. A Biol. Sci. Med. Sci. 60, 1553–1557.
van den Kroonenberg, A.J., Hayes, W.C., McMahon, T.A., 1996. Hip impact velocities
and body configurations for voluntary falls from standing height. J. Biomech.
29, 807–811.
Yang, Y., Schonnop, R., Feldman, F., Robinovitch, S.N., 2013. Development and
validation of a questionnaire for analyzing real-life falls in long-term care
captured on video. BMC Geriatr. 13, 40.