You are on page 1of 9

A HYBRID MODEL FOR PEDESTRIAN

IMPACT AND PROJECTION


DP Wood and CK Simms
Denis Wood Associates

Abstract
Real life accidents provide important information for understanding pedestrian/vehicle collisions. Collision speed is the
critical parameter in determining pedestrian injury severity and, where possible, this has traditionally been estimated from
tyre skid marks. However, the introduction of ABS brakes has increased the importance of pedestrian projection distance as a
means of estimating collision speed in real life accidents. This paper presents a theoretical Hybrid Model relating collision
velocity to pedestrian projection distance. The model combines the flight trajectory of a simple particle with the distribution
of post head-impact velocities predicted by Woods Single Segment Model (SSM). Comparison with staged and real life
accidents shows good predictions of minimum, maximum and mean projection distances. The overall distribution of collision
speeds has been used to predict upper and lower limits and a mean value of collision velocity for a given pedestrian
projection distance.

NOTATION

A lumped parameter relating collision velocity to
projection distance
c.g. centre of gravity
d
o
pedestrian c.g. height above the ground after
head impact
g gravitational constant
coefficient of friction between pedestrian and
the ground
M
P
pedestrian mass



M
V
vehicle mass
S
T
pedestrian projection distance
S
1-2
distance travelled by pedestrian c.g.
between leg and head impact.
V
H
horizontal velocity of pedestrian c.g.
after head impact
V
V
vertical velocity of pedestrian c.g.
after head impact
V
COL
collision velocity

INTRODUCTION

Analysis of real accidents is important for the understanding of pedestrian/vehicle collisions. A major
goal is to relate impact severity to injury severity: vehicle design can then be improved to reduce the
injury causation potential. Similarly, traffic management policy can be revised to reduce
pedestrian/vehicle collisions. The critical parameter in determining pedestrian injury severity is
collision speed and a procedure for estimating collision speed in real life accidents is therefore an
essential tool in crashworthiness research. The majority of pedestrian impacts are with the fronts of
vehicles. In these circumstances, tyre skid mark measurements have traditionally been the preferred
method of collision speed estimation. However, the introduction of ABS brakes has resulted in the
increasing importance of pedestrian projection distance for estimating collision speed. The strong
correlation between the square root of pedestrian projection distance and collision velocity predicted
by fundamental theory has been borne out by staged tests, and projection distance is therefore the
primary indicator of collision severity. This paper presents a theoretical model relating pedestrian
impact and projection distance to collision speed. The results of this model are compared with data
from staged and real life accidents.

MECHANICS OF PEDESTRIAN IMPACT

Overall Characteristics

An understanding of the mechanics of pedestrian/vehicle collisions has been the result of film
evidence of real and staged accidents. The former are chance recordings by traffic cameras in urban
areas. These are often poor quality but relate exactly what occurs in a real accident. Staged accidents
are usually captured on high speed video for clarity, and these tests feature cadavers and dummies as
human surrogates. These sources indicate that the primary movements of pedestrians in frontal
impacts follow definite patterns, but that there is variability in the secondary details. Frontal collisions
are categorised by the height of the impact location relative to the pedestrians centre of gravity (c.g.):

1. c.g. below the lower front edge of the vehicle.
2. c.g. between the upper and lower edges of the front of the vehicle.
3. c.g. above the upper front edge of the vehicle.

The last category represents the majority of adult/vehicle collisions and is the most important. In these
collisions, the front bumper strikes the lower leg region. The response to this depends on the height of
the bumper relative to the knee [1], and on the coefficient of friction between the feet and the ground.
The legs are accelerated in the direction of vehicle travel, while the upper body and head are brought
into contact with the bonnet and/or windshield.
There are therefore three distinct impacts. The first is between the lower leg and the bumper.
This is immediately followed by contact between the thigh area and the upper edge of the bonnet.
These impacts cause rotation of the upper body, resulting in a further impact between the
head/shoulders and the bonnet/windshield region. This reduces the angular velocity of the pedestrian,
but increases the (upward) vertical and horizontal velocity in the direction of vehicle motion.
Subsequent motion depends on collision speed. At low speeds, the pedestrian remains on the bonnet.
As impact speed increases, the increased rotational momentum causes the legs to rotate above the head
before falling back onto the bonnet. Braking of the vehicle usually causes the pedestrian to slide off
the bonnet. At yet higher impact speeds, the pedestrian somersaults onto the windshield/roof after head
impact. Again, vehicle braking causes the pedestrian to slide off the vehicle. Further increases in
impact speed can cause the pedestrian to pass fully over the vehicle.
There are, however, several factors that can cause pedestrian kinematics to diverge from the
description given. Variations in the front profile of the vehicle and in pedestrian height alter the head
impact location. Further, the severity of head impact is reduced if the shoulders strike the
bonnet/windshield prior to the head. Pre-impact transverse velocity of the pedestrian often results in
the pedestrian being projected to either side of the vehicle. Film evidence of staged collisions shows
that nominally equal impacts can cause the pedestrian to be deflected off the side of the vehicle or be
carried a considerable distance on the vehicle. In addition, flailing of limbs and rotation about the
pedestrians longitudinal axis introduce further variability.
Projection of the pedestrian from the vehicle is followed by a flight trajectory. This leads to
subsequent impacts with the ground as the pedestrian slides/rolls/bounces to rest. The majority of
injuries arising from these vehicle and ground impacts are trauma to the head, legs and pelvis. The
upper and lower legs and knees are most frequently injured which can lead to long-term disabilities,
but head injuries are generally more life threatening.

Modelling of Pedestrian Collisions

Significant advances in modelling pedestrian/vehicle collisions were achieved with the development
of multi-segment models of the human body followed by sophisticated software packages such as
MADYMO in the early 1980s. Investigations have included the effects of varying bonnet edge
location [2], predictions of head impact location [3] and head impact velocity [4]. Models with fewer
segments are frequently more stable than their more complex counterparts, with only a small loss in
accuracy [4]. Nevertheless, these deterministic models have not succeeded in predicting the variability
of head impact velocities that have been found under identical experimental conditions [5].
Woods single segment model (SSM) of an adult pedestrian impacting the front of a car was a
return to a simplified approach to pedestrian collisions [6]. This model predicted

multiple impacts between the pedestrian and vehicle.
similar head contact positions compared to experimental results.
head impact velocities of the same order of magnitude as experimental results.
times to head contact and ratios of wrap around ratio to pedestrian height comparable
to experimental data.

It can therefore be considered that Woods SSM model provides a good representation of the primary
pedestrian/vehicle interactions in frontal collisions.
There has been no emphasis on modelling pedestrian kinematics following head impact with
the bonnet/windshield. Existing models are based on elementary mechanics and rely on the
assumption that there is no further interaction between the pedestrian and the vehicle after head
impact. This provides the foundation for Searles particle model of pedestrian flight and subsequent
slide and bounce along the ground [7]. This model accounts for the loss of horizontal momentum due
to vertical impacts with the ground and allows multiple bounce of the pedestrian before coming to rest.
Searle [8] successfully validated his model using laboratory tests. However, it cannot be directly
validated against crash data, as the vertical and horizontal flight velocities of the pedestrian c.g. after
head impact (V
V
and V
H
) are not generally known.
Wood [9] combined predictions of (V
V
and V
H
) provided by his single segment model (SSM)
[6] with Searles trajectory model to estimate pedestrian projection distance (S
T
):

2 1
2
2
2
1

+ +
(

+
= S d
g
V
V
V
S
o
H
V
H
T

[1]

In equation 1, V
V
, V
H
and d
o
are derived using the SSM for a specific impact configuration, vis a vis
pedestrian attitude and anthropometrics, and vehicle mass, shape and braking rate, etc. A comparison
of equation 1 to experimental data published by the KOB project [10] gives good predictions of
projection distance in all cases, and figure 1 shows examples of this for a VW Passat (a) and a
Mercedes 200 (b).












0 5 10 15 20 25
0
5
10
15
20
projection distance - m
c
o
l
l
i
s
i
o
n

v
e
l
o
c
i
t
y

-

m
s
-
1
0 5 10 15 20
0
5
10
15
20
projection distance - m
c
o
l
l
i
s
i
o
n

v
e
l
o
c
i
t
y

-

m
s
-
1
(a) (b)

Figure 1. Collision velocity versus pedestrian projection distance:
experimental (circles) and Woods SSM (solid line) [9].

HYBRID MODEL

Theory

The assumption of instantaneous central (non-eccentric) impact of the pedestrian results in higher V
H

values than occur in practice. The consequent increase in projection distance is counterbalanced by
dropping the d
o
and S
1-2
terms in equation 1. This yields the following simplified or Hybrid model:

T COL
S A V = [2]


where
| |
(
(

|
|
.
|

\
|
+
(

+
H
V
V
P V
V
V
g
M
M M

1
2 2
1

= A

Model Requirements

In practice, specific values of the individual variables incorporated in the lumped parameter A in
equation 2 are either unknown or are subject to substantial statistical scatter. The Hybrid model is
therefore best applied using a statistical approach, and the distributions of the parameters M
V
,

M
P
,

,
V
V
/V
H
and

V
COL
as detailed in equation 2 are needed.

Vehicle mass
Analysis of DOT data [11] for the European car population shows that the distribution of vehicle mass
is normal with a mean of 932kg, and a standard deviation of = 217kg. Upper and lower bounds were
fixed at 1930kg and 585kg respectively.

Pedestrian mass
Pheasant [11] found the distribution of body weights to be normally distributed, and published data for
selected adult populations. Combining the British, French, USA, German and Swedish data for both
males and females gives a mean body mass of 68.37kg and a standard deviation of = 16.13kg. Upper
and lower bounds for pedestrian mass were fixed at the 3 limit.

Coefficient of pedestrian to ground friction
Wood and Simms [13] have reported a coefficient of friction with a normal distribution based on the
average value reported by ten independent research groups. The mean value for is 0.582, with a
standard deviation of = 0.1. Upper and lower bounds for were fixed at the 3 limit.

Ratio V
v
/V
h
after head impact
The SSM model was used to find the distribution of V
V
/V
H
for a range of vehicle types, braking
conditions and pedestrian anthropometrics. Simulations were performed with four different vehicle
types, ranging from a low wedge-shaped sports car to a flat fronted pontoon shape vehicle. For each of
these, the braking rate was varied from zero to maximum (
car
= 1.1g), and pedestrian size was varied
from 5% female to 95% male. Pedestrian attitude was varied to allow for front/back as well as side-on
collision with the vehicle. In addition, the simulations accounted for the possibility of bonnet droop
resulting from vehicle braking. Figure 2a shows the relationship between collision speed (V
COL
) and
V
V
/V
H
: flat fronted vehicle, 5% female side impact with moderate braking (
car
= 0.55g).
The ratio of V
V
/V
H
was determined for a total of 1132 simulations at four impact velocities:
11.2ms
-1
, 13.42ms
-1
, 17.92ms
-1
and 22.32ms
-1
. All configurations showed that above a critical collision
speed, the predicted ratio of V
V
/V
H
was positive. Consideration of pedestrian/vehicle interaction after
head-impact shows that negative values of V
V
/V
H
will result in immediate further body contact with the
bonnet. The ultimate effect of this is the reduction of V
V
/V
H
to zero. The mean of V
V
/V
H
at each speed
was therefore used to perform a power regression of V
V
/V
H
versus V
COL

|
0691 . 0
303 . 0
OFF COL
H
V
V V
V
V
= | [3]

where V
OFF
(the x-axis intercept) is the critical velocity below which V
V
/V
H
was set to zero. V
OFF
was
chosen to maximise the correlation coefficient (r = 0.997). Upper and lower bounds were fixed using
similar power regressions of the maxima/minima at each of the four speeds respectively.







0.


m







5 10 15 20 25
0
0.1
0.2
Impact velocity - ms
-1
V
V
/
V
H

f
r
o
m

S
S
M
0 5 10 15 20 25
0
0.2
0.4
6
Impact velocity - ms
-1
V
V
/
V
H

f
r
o

S
S
M
Mean
Min
Max

(a) (b)

Figure 2. (a) Individual V
COL
V
V
/V
H
relationship predicted by SSM; (b) Regression of
V
V
/V
H
against impact velocity.

Collision velocity distribution
For validation of the model, the collision velocity distribution of the appropriate test data has been
used. However, the application of the Hybrid model to pedestrian collisions in general requires
knowledge of the corresponding collision velocity distribution. Danner & Langwieder [14]
investigated the cumulative frequency of collision velocities (V
COL
) for 1353 pedestrian collisions and
reported 50% 4.263 ms
-1
and 90% 12.37 ms
-1
. This was used to model the general distribution of
V
COL
as log-normal, see figure 3.

0 5 10 15
0
0.2
0.4
0.6
0.8
1
collision velocity - ms
-1
c
u
m
u
l
a
t
i
v
e

f
r
e
q
u
e
n
c
y












Figure 3. Cumulative frequency of collision velocity: Danner&Langwieder (triangles)
and log-normal model (circles).

Monte-Carlo simulation
The distribution of A values for the Hybrid model was determined using Monte-Carlo simulation: a
value for V
COL
was chosen randomly from the distribution of V
COL
. The corresponding value of V
V
/V
H

was found from the regression in equation 3. Values for M
V
, M
P
and were chosen at random from
their respective normal distributions. A single value for A could then be computed. This process was
repeated with a loop-size of 20000 to yield the distribution of A

values.

VALIDATION OF HYBRID MODEL

Comparison with experimental data

The validity of the Hybrid model is best judged by comparison to experimental data of vehicle
velocity and pedestrian projection distance. The experimental data in this analysis is divided into two
categories: data collected from real accidents (N = 84, [10,15,16,17,18,19]), and data from staged tests
(N = 104, [10, 20, 21, 22, 23]). The latter category consists mainly of dummy tests (N = 85), but also
includes cadaver tests (N = 19). Experimental A values for both categories were found from

T COL
S A V = [4]

Tables 1&2 show the results of the Hybrid Model compared to experimental A values for real
accidents and staged tests respectively. The predicted maxima and minima are the upper and lower
0.25% limits respectively for the Hybrid model, whereas for the test data the actual maxima and
minima are presented. (The 0.25% limits were chosen to ensure upper and lower bound predictions
giving a high degree of confidence.) The distributions of V
COL
used in these predictions were the actual
distributions of V
COL
from the real accidents and staged tests subsets. This permits a direct validation of
the model.

Table 1. A values
Min Max Mean N
Hybrid Model
(V
COL
: real accidents)
2.34 4.30 3.12 _ 0.36
Real accidents 2.50 4.40 3.49 84 0.38

Table 2. A values
Min Max Mean N
Hybrid Model
(V
COL
: staged tests)
2.37 4.36 3.22 _ 0.39
Staged tests 2.53 3.90 3.23 104 0.26

Figure 4a&b shows the relationship between V
COL
and the square root of pedestrian projection distance
for real accidents (a) and staged tests (b). The Hybrid model mimimum, maximum and mean collision
velocity predictions based on the actual V
COL
distributions in each case are also shown.



Ma
50
0 2 4 6 8 10
0
10
20
30
40
50
square root projection distance m
1/2
c
o
l
l
i
s
i
o
n

v
e
l
o
c
i
t
y

m
s
-
1
Min
Mean
x
0 2 4 6 8 10
0
10
20
30
40
square root projection distance m
1/2
c
o
l
l
i
s
i
o
n

v
e
l
o
c
i
t
y

m
s
-
1
Min
Mean
Max









(a) (b)

Figure 4. Collision velocity versus square root of projection distance for (a) real accidents and
(b) staged tests. Maximum, minimum and mean hybrid model predictions based the
actual V
COL
distributions in each case are also shown.

Model predictions for pedestrian collision distribution

It was necessary to use the distributions of V
COL
from staged tests and real accidents for validation
purposes. However, in developing an overall model for all pedestrian impacts, the log-normal
distribution of V
COL
based on the data reported by Danner & Langwieder provides a more accurate
representation of the actual distribution of collision velocities. This is a far larger sample and
represents all pedestrian accidents. The predictions of the Hybrid model using this distribution of V
COL

are compared with the results of real accidents and staged tests, see table 3. Graphic representations of
this information are given in figures 5&6.

Table 3. A values
Min Max Mean
Hybrid Model
(V
COL
: Danner & Langwieder)
2.51 4.47 3.55
Real accidents 2.50 4.40 3.49
Staged tests 2.53 3.90 3.23


DISCUSSION

Comparison of the predictions from the Hybrid Model with the experimental (staged and real life)
collisions is shown in figure 4a&b, and in tables 1&2. All but two of the experimental tests lie within
the limit bounds predicted by the Hybrid model. For real accidents (table 1 & fig 4a), the predicted
limits match the experimental upper and lower bounds very closely, while the predicted mean is
somewhat lower than the experimental mean. For staged tests (table 2 & fig 4b), the predicted mean
value is very close, while the Hybrid model predicts more scatter than was found experimentally. The
low scatter in the staged tests can be partially due to the predominance of 50
th
percentile male
dummies used in staged collisions.
The higher predicted mean for the staged tests compared to the predicted mean in real
accidents is explained by consideration of the regression of V
COL
against V
V
/V
H
(equation 3) and the
mean collision velocity of real accidents (15.36ms
-1
) compared to staged tests (12.54m
s-1
). Figure 2
shows that below a critical collision velocity (V
COL
= 10.6ms
-1
) the mean value of V
V
/V
H
is zero: the
distribution of staged tests V
COL
is clearly biased towards yielding V
V
/V
H
= 0, and the A values
calculated using equation 2 will therefore also be higher.


















0 2 4 6 8 10
0
10
20
30
40
50
square root projection distance m
1/2
c
o
l
l
i
s
i
o
n

v
e
l
o
c
i
t
y

m
s
-
1
Min
Mean
Max




Figure 5. Experimental data and model predictions.














2.2 2.6 3 3.4 3.8 4.2 4.6
0
0.5
1
A value
c
u
m
u
l
a
t
i
v
e

f
r
e
q
u
e
n
c
y


Figure 6. Cumulative frequencies of A values: staged tests (circles), real accidents (squares) and
Hybrid Model (triangles).

The use of the Danner & Langwieder collision speed distribution (V
COL
) increases the predicted mean
value of A, because the preponderance of collision speeds are now below the cut-off velocity for the
V
V
/V
H
regression (see equation 3 and figure 2). The Hybrid model does not account for continued post-
impact head interaction with the vehicle above the critical collision velocity (V
OFF
in equation 3). Film
analysis shows that the effect of this interaction is to reduce the ratio of V
V
/V
H
at eventual separation
relative to its value immediately after head impact. This reduction results in a corresponding increase
in A (see figure 2), similar to the increase in A value due to the continued pedestrian/vehicle
interaction found in practice. This explains the excellent match between real accidents and the Hybrid
model predictions shown in figure 6. The following equations are therefore proposed for estimation of
collision velocity from pedestrian projection distance:

Minimum
T COL
S V 5 . 2 = [5]

Mean
T COL
S V 6 . 3 = [6]

Maximum
T COL
S V 5 . 4 = . [7]


CONCLUSIONS

A Hybrid model of pedestrian to vehicle impact has been developed to relate collision velocity to
projection distance. The Hybrid model has a theoretical basis with limits determined by the statistical
distributions of its component variables. These resulting predictions are theoretical rather than
empirical, and hence are not based on any of the experimental data used in the validation. Comparison
with staged and real life accidents shows very good predictions of minimum, maximum and mean
projection distances. The overall distribution of collision speeds has been used to predict upper and
lower limits and a mean value of collision velocity for a given pedestrian projection distance.






REFERENCES

1. Wood D.P., Mathematical analysis of pedestrian lower leg impacts, Proc. IRCOBI Conf, pp 169-182, 1989.
2. Pritz H. B. & Pereira J. M., Pedestrian hip impact simulator development and hood edge location consideration on injury
severity, SAE No. 831627, 1983.
3. Ashton S. J., Cesari D. & van Wijk J., Experimental reconstruction and mathematical modelling of real world pedestrian
accidents, SAE No. 830189, 1983.
4. Van Wijk J., Wismans J., Maltha J. & Wittebrood L., MADYMO pedestrian simulations, SAE No. 830060, 1983.
5. Cavallero C., Cesari D., Ramet M., Billault P., Farisse J., Seriat-Gautier B. & Bonnoit J., Improvement of pedestrian
safety: Influence of shape of passenger car front structures upon pedestrian kinematics and injuries: evaluation based on
50 cadaver tests, SAE No. 830624, 1983.
6. Wood D.P., Impact and movement of pedestrians in frontal collisions with vehicles, Proceedings Inst. Mech. Engnrs,
Vol 202 No. D2, p 51, 1988.
7. Searle J.A. & Searle A., The trajectories of pedestrians, motorcycles, motorcyclists, etc., following a road accident, SAE
No. 831622, 1983.
8. Searle J. A., The physics of throw distance in accident reconstruction, SAE No. 930659, 1993.
9. Wood D.P., Application of a pedestrian impact model to the determination of impact speed, SAE No. 910814, 1991.
10. Cesari D., Personal Communication giving details of pedestrian impact tests under the KOB project, 1989.
11. Department of Transport, Transport statistics report. Cars: make and model. Injury, accident and casualty rates, Great
Britain, 1991, Publ London HMSO, 1993.
12. Pheasant S., Bodyspace : anthropometry, ergonomics and design, Taylor and Francis, London and Philadelphia, p 120,
1986.
13. Wood D.P & Simms C.K., Coefficient of friction in pedestrian throw, Impact, Vol 9, no 1, pp 12-15, 2000.
14. Danner M. & Langwieder K., Collision characteristics and injuries to pedestrians, Proc 7
th
ESV Conference, Paris, 1979.
15. Eubanks J., Pedestrian involved traffic collision reconstruction methodology, SAE No. 921591, 1992.
16. Eubanks J., Pedestrian tests reported in Accident Reconstruction Journal, vol 5, no. 6, pp 30-32, 1993.
17. Severy D. & Brink H., Auto-pedestrian collision experiments, SAE No. 660080, 1966.
18. Open Day Tests, Institute of Traffic Accident Investigators, Leyland, Lancashire, England, 1993.
19. Steffan H., Moser A., B.C. Geigl & Motomiya Y., Validation of the coupled PC-Crash MADYMO occupant
simulation model, SAE No. 2000-01-0471, 2000.
20. Grandel J., Zeisberger H. & Walz F. H., Kinematics and head injuries in vehicle/pedestrian accidents at speeds above 50
kph, Proc. IRCOBI Conf., pp 189-204,1986
21. Hill G. S. Calculations of vehicle speed from pedestrian throw, Impact, pp 18-20, Spring 1994.
22. Dettinger J., Methods of improving the reconstruction of pedestrian accidents: development differential, impact factor,
longitudinal forward trajectory, position of glass splinters (in German), Verkehrsunfall und Fahrzeugtechnik, December
1996, 324-330; January 1997, 25-30 (two parts).
23. Schneider H. & Beier G., Experiment and accident: Comparison of dummy test results and real pedestrian accidents,
SAE No. 741177, 1974.0

You might also like