Accident Analysis and Prevention 111 (2018) 311–320

Contents lists available at ScienceDirect

Accident Analysis and Prevention

journal homepage: www.elsevier.com/locate/aap

Real life safety benefits of increasing brake deceleration in car-to-pedestrian T

accidents: Simulation of Vacuum Emergency Braking
⁎
Hanna Jeppsson , Martin Östling, Nils Lubbe
Autoliv Research, Wallentinsvägen 22, 447 83, Vårgårda, Sweden

A R T I C L E I N F O A B S T R A C T

Keywords: The objective of this study is to predict the real-life benefits, namely the number of injuries avoided rather than
Automated emergency braking the reduction in impact speed, offered by a Vacuum Emergency Brake (VEB) added to a pedestrian automated
Pedestrian emergency braking (AEB) system. We achieve this through the virtual simulation of simplified mathematical
Effectiveness models of a system which incorporates expected future advances in technology, such as a wide sensor field of
Risk curves
view, and reductions in the time needed for detection, classification, and brake pressure build up.
The German In-Depth Accident Study database and the related Pre Crash Matrix, both released in the be-
ginning of 2016, were used for this study and resulted in a final sample of 526 collisions between passenger car
fronts and pedestrians. Weight factors were calculated for both simulation model and injury risk curves to make
the data representative of Germany as a whole. The accident data was used with a hypothetical AEB system in a
simulation model, and injury risk was calculated from the new impact speed using injury risk curves to generate
new situations using real accidents.
Adding a VEB to a car with pedestrian AEB decreased pedestrian casualties by an additional 8–22%, de-
pending on system setting and injury level, over the AEB-only system. The overall decrease in fatalities was
80–87%, an improvement of 8%. Collision avoidance was improved by 14–28%.
VEB with a maximum deceleration in the middle of the modelled performance range has an effectiveness
similar to that of an “early activation” system, where the AEB is triggered as early as 2 s before collision. VEB
may therefore offer a substantial increase in performance without increasing false positive rates, which earlier
AEB activation does.
Most collisions and injuries can be avoided when AEB is supplemented by the high performance VEB; re-
maining cases are characterised by high pedestrian walking speed and late visibility due to view obstructions.
VEB is effective in all analysed accident scenarios.

1. Introduction
Pedestrian fatalities and injuries are frequent: according to the
World Health Organisation (2015), 270,000 pedestrian fatalities ac-
count for 22% of a total 1.25 million road traffic fatalities. This fre-
quently cited report might underestimate the true size of the problem.
Bhalla et al. (2014) for example estimate that 460,000 pedestrian
fatalities account for 35% of a total 1.33 million road traffic fatalities.
In the European Union (EU), 5621 pedestrian fatalities comprised 21%
of all road fatalities in 2014 (European Commission, 2016). In Ger-
many, 537 pedestrian fatalities (16% of total road traffic fatalities) and
31,073 non-fatal pedestrian casualties (8% of total non-fatal road traffic
casualties) were reported for 2015 (Bundestag, 2016).
Better protection of pedestrians in road traffic has been called for,
specifically through reduced driving speeds, improved vehicle design
with more forgiving car fronts, and greater traffic flow separation
(World Health Organisation, 2015). EU initiatives to improve road
safety (European Commission, 2010) resulted in an overall reduction in
fatalities of 22% between 2010 and 2013, but of only 11% for pedes-
trians. Further actions to improve the situation were proposed, such as
encouraging safe roadway infrastructure and assessing the effectiveness
of emergency brake systems with pedestrian detection (European
Commission, 2015). Germany reports figures in line with those of the
EU, with a below-average fatality reduction for pedestrians in the mid-
term evaluation of the 2011–2020 targets, and has called for a greater
focus on pedestrian safety (BMVI, 2015).
The EU is advanced in pedestrian protection. The European
Enhanced Vehicle-Safety Committee (EEVC) published test methods
and requirements for pedestrian protection in the 1990s (European
Enhanced Vehicle-Safety Committee, 2002). These formed the basis for

⁎
Corresponding author.
E-mail address: hanna.jeppsson@autoliv.com (H. Jeppsson).

https://doi.org/10.1016/j.aap.2017.12.001
Received 28 April 2017; Received in revised form 1 December 2017; Accepted 1 December 2017
Available online 17 December 2017
0001-4575/ © 2017 Elsevier Ltd. All rights reserved.
H. Jeppsson et al. Accident Analysis and Prevention 111 (2018) 311–320

the pedestrian protection measures mandated for new vehicle types in

the EU since 2005, which require vehicle manufacturers to comply with
a set of tests to ensure forgiving car fronts in case of collision (EC di-
rective 2003/102). The European New Car Assessment Program (Euro
NCAP) has tested and published results using similar but more stringent
tests since 1997 to provide consumers with a safety rating (van
Ratingen et al., 2016). These ratings have been shown to correlate with
real life benefits: higher rated vehicles have been observed to cause
fewer pedestrian injuries (Strandroth et al., 2011; Pastor, 2013).
Euro NCAP has recently gone further, introducing in January 2016 a
rating for automated emergency braking (AEB) for pedestrian protec-
tion which evaluates a car's ability to reduce speed to mitigate or avoid
collisions with pedestrians crossing in front of it (Schram et al., 2015).
Pedestrian AEB is predicted to have great potential in reducing pedes-
trian casualties, both when implemented on typical levels of car-front-
forgiveness (Rosén et al., 2010; Rosén, 2013; Lindman et al., 2010) and
when used in combination with pedestrian airbags (Fredriksson and
Rosén, 2012; Edwards et al., 2015).
AEB is more effective in reducing speed the more reliably it acti-
vates (dependent upon sensor accuracy and cover area, time delays in
detection, and classification of event), the earlier it activates (depen-
dent upon sensor and system delays), and the harder it decelerates
(dependent upon brake activation delay, brake pressure build-up, and
maximum deceleration).
The relationship between these parameters and estimates of AEB
effectiveness has recently been explored. Edwards et al. (2014) quan-
tified the increased monetary benefits in Germany and the UK of three
hypothetical AEB systems, representing future advances in technology,
when time delays for detection, classification, and brake pressure build
up, were decreased. A maximum deceleration of 10 m/s2 for dry road
and 8 m/s2 for wet roads was used for all systems. Rosén et al. (2010)
showed that AEB effectiveness in reducing severe and fatal injuries
increased when sensor field of view (FoV) was increased, when max-
imum brake deceleration was increased from 2 m/s2 up to 10 m/s2, and
when brake activation delay was reduced. Rosén (2013) quantifies the
expected real life consequences of further system limitations such as
deactivation of the AEB system above 60 km/h driving speeds and
during night time. However, none of these studies investigated the
benefits of increasing maximum brake deceleration to levels above
10 m/s2, possibly because it seemed unlikely that such deceleration
levels could be achieved and sustained.
Increasing the rate of deceleration would be expected to deliver
significant safety benefits, but achieving this is challenging. A brake
airbag has been proposed which can increase brake deceleration levels
from 10 m/s2 to 20 m/s2 for up to 0.1 s, after which deceleration levels
drop below 10 m/s2 (Mellinghoff et al., 2009). Overall, it takes longer
to come to a standstill with the brake airbag than with continued
conventional braking at 10 m/s2. Activation, therefore, needs to be
timed carefully and the safety benefit appears to be limited.
We here propose and model the safety benefits of a Vacuum
Emergency Brake (VEB), also called the Torricelli Brake (Lang, 2015).
The VEB can produce a sustained increase in deceleration. The new
system is described below, and the remainder of this paper presents
research to quantify the expected safety benefits of adding the VEB to
an AEB system.
The VEB consists of a vacuum tank with opening valve, a release
mechanism, linkage to the vehicle and rubber plate for the ground
contact (Fig. 1a). On activation, a release mechanism shoots the unit
towards the ground, and 0.07 s later, when the unit hits the ground, a
valve to the vacuum tank is opened. The atmospheric pressure presses
the unit toward the ground and the linkage transmits force both in the
driving direction (friction force) and upwards against the vehicle
(normal force). The force level is dependent on the plate area acting on
the ground, while the force duration is dependent on the rubber sealing
and the size of the vacuum tank. With a plate area of 0.36 m2, the peak
deceleration increased by a factor of 1.8 against baseline in tests with a
Fig. 1. (a) VEB components. (b) VEB installation in test.
road friction coefficient of 0.9. This translates to an increase of max-
imum vehicle deceleration from 9 m/s2 to 16 m/s2. The VEB is best
mounted behind but close to the center of gravity of the vehicle. Due to
conflicts with surrounding parts in the test vehicle, it was positioned
close to the rear wheel axis in our tests (Fig. 1b).
While it is clear that pedestrian protection will increase when using
VEB, as achievable speed reduction is always larger than that achiev-
able without VEB, the exact real-life benefit of a realistic and sustained
maximum deceleration of more than 10 m/s2 is uncertain.
The objective of this study, therefore, is to predict the number of
casualties avoided in accidents through virtual simulation of simplified
mathematical models of a VEB added to a pedestrian AEB system. The
system modelled incorporates future advances in technology, such as
short delays for detection and classification, a reduction in the time
needed for brake pressure build up (Edwards et al., 2014), and a wide
sensor field of view. An analysis of benefit sensitivity to road friction
levels and system settings, alongside a characterization of casualties not
prevented (remaining cases), concludes this analysis.
We denote this as “real-life benefits” as we predict the effect on
injuries and fatalities in the real world as opposed to “test benefits”
where one predicts the effect on impact speeds in a limited set of test
scenarios. One should take these estimates with due care, they are
approximations, and need to be confirmed with retrospective studies
once sufficient real-world data becomes available.
This paper is structured as follows: Section 2 reviews and discusses
existing methods to evaluate safety benefits of pedestrian AEB systems.
Section 3 details the methods employed in this study to quantify the
real-life benefit of VEB. Section 4 presents the results, and is followed
by discussion in Section 5 and the conclusion in Section 6. We limit the
description of the simulation model in Section 3.4 to the essentials;
details on system parameter settings are presented in Appendix A.
Appendix B gives confidence intervals for the injury risk curves which
are omitted from Fig. 4, Section 4.1 for clarity and brevity. Appendix C
presents a model fit assessment for the injury risk curves.
2. Review of existing safety benefit evaluation methods
Predicting real-life safety benefits of new vehicle safety technology
can be carried out in many ways. Here we review methods for the
prospective assessment of safety benefits offered by vehicle-based

312
H. Jeppsson et al.
active safety technology aimed at protecting pedestrians from vehicle-
to-pedestrian collisions.
System effectiveness can be estimated in a Field Operational Test
(FOT) whereby the new technology is installed in a vehicle fleet driven
on public roads, thus mimicking normal usage. The safety benefits in
these approximated normal driving conditions can then be estimated by
comparing collision events with and without the technology (Bärgman,
2015). However, such an evaluation of collision events necessitates an
extensive amount of data, typically several million to hundreds of
millions of driven kilometres, to get reliable results, which is often not
realistically feasible (Bärgman, 2015).
Instead, surrogate safety indicators, that is safety-critical events
defined by close proximity or severe braking, are often used to evaluate
safety benefits, as safety-critical events are more frequent than colli-
sions. However, the underlying assumption that surrogate safety in-
dicators can reliably predict crash occurrence is debated (Hydén, 1987;
Jonasson and Rootzén, 2014; Bärgman,2016).
Extreme Value Theory (Coles, 2001) can deliver a more formal re-
lation between safety-critical events and crashes while avoiding the
pitfalls of arbitrary definitions of a safety-critical event (Tarko, 2012).
Extreme Value Theory has been applied to predict crash occurrence
from non-crash events in FOTs (Jonasson and Rootzén, 2014) and
Driving Simulator studies (Farah and Azevedo, 2017), but not yet to
pedestrian protection evaluations.
FOTs are complex, require vehicle instrumentation and volunteers
to drive in traffic. The need for unrealistic amounts of data can be
balanced with applying approximations based on critical events rather
than crashes, but this inevitably introduces sources of bias and error. No
FOT evaluation aiming to quantify the benefit of pedestrian AEB system
has been undertaken to date.
Simulation methods are becoming increasingly popular, making the
evaluation of millions of km more affordable. A first differentiation of
simulation methods can be made between traffic simulation and single
accident simulation. Traffic simulation generates traffic flow including
accident and non-accident situations based on previously-collected in-
formation on traffic and accidents, replicating this in mathematical
models of the traffic environment, vehicle, and safety system (Helmer,
2014; Tanaka, 2015). Simulating hundreds of millions of km of driving
is fairly easy. However, collecting the data needed to accurately model
traffic with all the possible dependencies between different describing
variables and validating the mathematical models requires tremendous
effort (Sander and Lubbe, 2016).
Toyota developed a traffic simulation framework consisting of road
environment, vehicle, and driver behaviour models. Simulating driver
error in various traffic situations leads to collisions in the simulated
traffic (Tanaka, 2015). This simulation framework was applied to
evaluate the effect of pedestrian FCW and AEB in Japan (Morales
Teraoka et al., 2013) and Europe (Morales Teraoka and Tanaka, 2014).
However, effectiveness estimates are not presented.
BMW developed a traffic simulation tool to estimate ADAS benefits
for pedestrian safety called S.A.F.E.R. According to Kompass (2012),
S.A.F.E.R. simulates all relevant processes in detail to obtain re-
producible estimates of key indicators such as system effectiveness and
False Positive rates. Helmer (2014) has studied effectiveness and false
positive rates for pedestrian AEB systems. The quality of the data to
model traffic, which was a mix of German accident data and US normal
driving data, and the accuracy of representation of dependencies be-
tween variables, remain questionable.
Single accident simulation, on the other hand, makes use of data
collected on accidents and replicates pre-collision situations for models
of vehicles with and without the safety system under consideration
(Erbsmehl, 2009). By using complete accident records, dependencies
between variables are naturally being taken into account and so-called
surprise events are more likely to be captured (Sander and Lubbe,
2016).
Examples of single accident reconstruction applied to pedestrian
313
Accident Analysis and Prevention 111 (2018) 311–320
AEB include the VW group’s tool rateEFFECT (Wille and Zatloukal,
2012; Roth et al., 2017), work conducted at IFSSTAR (Hamdane et al.,
2015) and Autoliv’s estimate for pedestrian and cyclist AEB effective-
ness (Rosén, 2013). Hamdane et al. (2015) and Rosén (2013) used
simplified vehicle kinematics and ideal sensor models, which may lead
to an overestimation of effectiveness as real systems may not yet detect
pedestrians perfectly.
Simulation-based estimates for car-to-car rear end AEB effectiveness
in avoiding collisions of approximately 80% (Sander, 2017; Bärgman
et al., 2017) exceed estimates from retrospective analyses of effective-
ness in collision avoidance: 50% in police reported collisions (Cicchino,
2017) and 38% in injury collisions (Fildes et al., 2015). Bärgman et al.
(2017) argue that the differences are likely to originate from the dif-
ferences in the formulation of the system logic. The system logic in
Bärgman et al. (2017) and Sander (2017) is based on the deceleration
required, and is more effective than the TTC-based system logic em-
ployed in a simulation study by Kusano and Gabler (2012) which does
not initiate braking below 15 km/h. Kusano and Gabler’s effectiveness
estimate for injury reduction of AEB of 50% is close to the retrospective
estimates from Cicchino (2017) and Fildes et al. (2015). In contrast,
collision avoidance of 7.7% is far from the approximate 80% given by
Bärgman et al. (2017) and Sander (2017). Hence, it appears that the
performance differences between systems is far larger than the potential
inaccuracies introduced by simplified models such as ideal sensors in
simulations.
For pedestrian AEB, no retrospective effectiveness estimates of
production systems in the market are publically available as systems are
fairly new and data is sparse. Ohlin et al. (2017), for example, found
that only 31 vehicles involved in pedestrian and cyclist collisions in
Sweden from 2003 to March 2014 were equipped with pedestrian AEB,
and did not find statistically significant effects on injury collisions. It is
not yet possible, therefore, to quantify the effect that simplifying si-
mulation models has on performance estimation, but the conclusions
from car-to-car rear end AEB evaluations indicate that simulation will
yield useful results.
Attempts have been made to reduce the complexity and number of
types of different traffic encounters and sift out those that are most
representative and relevant to safety benefit analyses. Regulatory and
consumer testing in this field is commonly carried out with very few
hardware tests. Euro NCAP, for example, tests only 3 scenarios at dif-
ferent driving speeds for pedestrian encounters (Schram et al., 2015).
Edwards et al. (2014), Edwards et al. (2015), and Yanagisawa et al.
(2017) provide further examples of using a limited set of test scenarios
to predict system performance. While such attempts can be successful –
as demonstrated by the correlation between Euro NCAP passive safety
pedestrian score and real-world performance (Strandroth et al., 2011;
Pastor, 2013) – it remains yet to be seen whether any of the existing test
series provide accurate real-world estimates. Until then, simulating a
wide range of different accidents should at least complement perfor-
mance estimates based on these limited number of tests.
In conclusion, a simulation-based method was selected as most
suitable to predict the benefit of VEB.
3. Methods
German PCM data was used together with simplified models for
pedestrian AEB and VEB systems in virtual simulation to predict the
reduction in injuries achievable by these systems.
3.1. Data
The German In-Depth Accident Study, GIDAS, is a database which
collates details of accidents investigated in and around Dresden and
Hanover by specialist investigation teams. The teams are contacted if a
crash occurs within these regions and an injury is suspected by the
police. An extensive investigation is then undertaken to provide a
H. Jeppsson et al.
detailed perspective on the accident, including the sequence of events
and the cause (Erbsmehl, 2009). Although the GIDAS data is collected
with the aim of being representative of Germany as a whole, it over-
represents severe and fatal accidents. This can be compensated for by
weighting the data using German national road traffic accident statis-
tics.
The cases in the GIDAS-based Pre Crash Matrix, PCM, are selected
by the GIDAS to constitute a representative subset of the GIDAS data-
base. Crashes are reconstructed to generate a detailed description of the
trajectories (including velocities and accelerations) of both car and
vulnerable road user (VRU) during the final seconds before collision.
Based on the reconstruction, data is generated at a rate of 100 times per
second (100 Hz) for a time span beginning 5 s before the collision. In
the GIDAS PCM, both cars and VRUs are modelled as rectangular ob-
jects. Exact car dimensions are available in each case, while pedestrians
are given a size of 0.8 × 0.4 m (Erbsmehl, 2009).
GIDAS data released January 2016 and PCM data released February
2016 were used for this study from which a sample was selected as
follows. For the years 1999–2015, the GIDAS database contains 3548
accidents involving pedestrians with reported injury, of which 1385
also have additional information in the GIDAS PCM database.
Removing cases where the severity of injury was not recorded reduces
the number to 1384. Restricting the count to pedestrians struck by an
M1 vehicle, a vehicle not having more than eight seats in addition to
the driver's seat, decreases the number of cases to 1348. Of these, 766
cases involved vehicles with registration year 1997 (the year Euro
NCAP started publishing pedestrian ratings and thereby the car man-
ufacturer started to enhance the vehicle for better pedestrian safety) or
later. Since AEB is only effective in frontal impacts, a Matlab script then
analysed these cases to assess the degree to which the pedestrian col-
lided with the front of the vehicle compared to the side; only cases
where they collided with the front to a greater degree were kept in the
sample, reducing the number of accidents to 559. Finally, removing
cases where the Maximum Abbreviated Injury Scale (MAIS) level was
missing resulted in a final sample of 526 accidents.
Weight factors were calculated for both the simulation model and
injury risk curves to make the PCM data representative of Germany as a
whole. First, weight factors were calculated for all pedestrian casualties
(1384) recorded in the PCM database to replicate the distribution of
fatal, severe, and slight pedestrian injuries (raw weight factors).
Second, these factors were normalized to obtain an equal total number
(526) of cases for the unweighted and weighted samples (see Table 1).
3.2. Data characterization
The final weighted sample characterization is described in Fig. 2a–f.
Fig. 2a shows the Time to Collision (TTC) when the pedestrian becomes
visible to the car sensor. TTC is the predicted time to collision, assuming
an unbraked vehicle and a pedestrian in constant motion, when the
brake decision is taken. TTC is calculated using Eq. (1),
Stot
TTC =
v0 (1)
where v0 is the instantaneous speed in m/s and Stot is the available
distance to the predicted collision point in m.
Fig. 2b–d depicts the speed of the vehicle, the walking speed of the
Table 1
Weight factors for the PCM database.
Germany PCM
Numbers (%) Numbers (%)
Fatal 537 1.7 52 3.8
Severe 7792 24.7 690 49.9
Slight 23281 73.7 642 46.4
Σ 31610 100 1384 100
Accident Analysis and Prevention 111 (2018) 311–320
pedestrian at impact, and the road friction coefficient at the accident
scene. In Fig. 2e, the vehicle front is divided laterally into 10 fields,
showing where on the vehicle front the impact occurs together with
information as to which side the pedestrian entered the driving path
(from the right, the left, or from other, mainly longitudinal). Most
impacts occur at the leftmost 10% of the vehicle width, with the pe-
destrian entering from the left. The distribution of crash scenarios is
given in Fig. 2f; the most common impact scenarios are where the ve-
hicle is going straight and the pedestrian is crossing the road either
from the right (38.2%) or from the left (24.8%).
3.3. Injury risk curves
Pedestrian injury risk curves quantify the probability of sustaining
an injury dependent on collision speed. Risk curves for the three injury
severities fatal, MAIS3 + F and MAIS2 + F were calculated from PCM
data for a pedestrian struck by the front of a modern M1 vehicle (see
filtering method in 2.1), using normalized and weighted data (Table 1).
Logistic regression was performed using R glm function, with collision
speed as the independent variable, injury outcome at the considered
severity level as the dependent variable, and case weights according to
Table 1. Injury risk curves are defined in Eq. (2)
e b0 + b1 v
risk (v ) =
1 + e b0 + b1 v (2)
where v is the impact speed in km/h, and b0 and b1 are the regression
coefficients.
3.4. Simulation model
The PCM accident data was used with a hypothetical AEB system in
a simulation model, and an injury risk was calculated from the new
impact speed using the injury risk curve. This generates new situations
using real accidents. The framework of the Matlab implementation to
simulate PCM cases was adopted from Rosén (2013).
The functionality of the AEB system was described by a number of
system parameters. Those were set to reasonable values for future AEB
systems. All system parameters can be found in Appendix A.
The AEB system was activated if, and only if, the following five
requirements were satisfied (Rosén, 2013) (Fig. 3):
1. VRU visible (visibility was calculated for all four pedestrian corners)
and within sensor FoV and range during time for classification.
2. VRU within trigger width which is the lateral extension from the
driving path (one corner minimum).
3. Predicted TTC (for the car when it starts braking) less than TTClimit.
TTClimit is the earliest Time to Collision when the system is allowed
to activate. The system might activate later, at lower TTC values,
depending on other conditions, but not earlier, at TTC values ex-
ceeding TTClimit. See Table 2 for parameter settings of TTClimit.
4. Car predicted to collide with VRU.
5. Required distance to come to full stop smaller than available dis-
tance with 1 m safety margin.
Six different systems are here evaluated (Table 2). The reference
system used adopted parameters similar to those of Rosén (2013)
Sample numbers (%) Raw weight factors Normalised factors
29 5.5 1 0.48
278 52.9 1.09 0.53
219 41.6 3.51 1.68
526 100
314
H. Jeppsson et al. Accident Analysis and Prevention 111 (2018) 311–320

Fig. 2. (a) Cumulative distribution of Time to

Collision at Point of Visibility. (b) Cumulative
distribution of vehicle impact speed. (c)
Cumulative distribution of pedestrian impact
speed. (d) Cumulative distribution of friction
coefficient. (e) Distribution of frontal impact on
vehicle front. (f) Distribution of accident sce-
narios.

315
H. Jeppsson et al. Accident Analysis and Prevention 111 (2018) 311–320

Table 3
Result from logistic regression.

Best estimate Lower limit Upper limit

Fatal b0 −7.04 −5.38 −8.69

b1 0.09 0.12 0.06

MAIS3+F b0 −5.07 −4.16 −5.98

b1 0.09 0.11 0.07

MAIS2+F b0 −2.48 −1.99 −2.97

b1 0.06 0.08 0.05

Fig. 3. Description of decision algorithm.

except that FoV was set to 180° and no minimum range for detection
was used. The maximum deceleration of the AEB system, amax, was
1 g*μ where μ was the coefficient of friction from PCM (recall Fig. 2d)
and g the gravitational constant (9.81 m/s2). For the VEB systems an-
other amax was used based on the dimension of the VEB. Three variants
were evaluated representing three different sizes of the VEB with the
smallest having dimensions 0.4 × 0.4 m and the largest 0.6 × 0.6 m
(see Table 2 for amax levels). For example “VEB high” has an amax of
1.8*g*μ; this formula is valid for all road conditions.
Two additional system settings were simulated for comparison. The
first, “Perfect road”, assumed that a maximum deceleration of 9 m/s2
could always be reached, which would be the case if all roads were of
the standard of the best roads, as indicated by the maximum value in
Fig. 2d. The second, “Early activation”, assumed the same maximum
Fig. 4. Pedestrian injury risk curves for modern M1 vehicle frontal accidents.
deceleration as the reference system, but activated earlier with a
greater trigger width (Table 2).
System effectiveness — the percentage reduction of fatal, MAIS3+F terms of impact speed reduction, but pedestrian injury reduction.
and MAIS2+F injuries — was assessed using a standard dose-response Regression coefficients for the injury risk curves as defined in Eq. (2)
method (Korner, 1989; Kullgren, 2008; Fredriksson and Rosén, 2012). are given in Table 3. The exponential function of the coefficient b1 gives
The number injured without system was obtained by summing up the odds ratio. For fatal and MAIS3+F the odds ratio is the same, 1.09,
weighted injury risks at the original collision speeds for each impact but for MAIS2+F the value is smaller, 1.06. This means that the curve
using injury risk curves. The combined AEB and VEB systems resulted for MAIS2+F is flatter and for each 1 km/h collision speed increase, the
in a reduction of impact speed. The same injury risk curves were then odds of sustaining injury increase by 6%, whereas for MAIS3+F and
used to calculate the risks of being injured with these new impact fatal injuries the odds increase by 9%. The injury risk curves are shown
speeds. The weighted (for the original injury outcome of the case being in Fig. 4. Injury risk curves for fatal, MAIS3+F and MAIS2+F with 95%
slight, sever or fatal) sum of these risks gave the number injured with confidence intervals are shown in Appendix B.
the system. The effectiveness was defined in Eq. (3) as the relative
difference between the numbers injured at the original speeds and the
modified speeds. 4.2. Real world benefit

N − N' N′ ⎞ The real-world benefit calculated as the effectiveness of the systems

E= *100% = ⎛1 − *100%
N ⎝ N⎠ (3) (Eq. (3)) is shown in Table 4. The pedestrian AEB system with VEB
increases effectiveness by between 8% and 22% depending on which
For example, a system with an effectiveness of 10% for MAI2+F
system setting and injury level is compared. The level of avoidance
injuries is preventing 10% of all MAIS2+F injuries.
increases by between 14% and 28% compared to the “Reference” AEB.
With the lowest performance setting of VEB, the effectiveness matches
4. Results that of “Perfect road”. “VEB mid” performance gives a similar effec-
tiveness to that of the “Early activation” system, where trigger width
4.1. Injury risk curves and TTC were increased to 2 m and 2 s, respectively. Finally, the “VEB
high” setting gives effectiveness values approaching 90% and an
The injury risk curves are used to relate impact speeds to injury avoidance level above 80%. An analysis of vehicle impact speed
outcome. This allows the benefits of VEB to be quantified not only in (Fig. 7c) shows why effectiveness was so high for “VEB high”: very few

Table 2
Summary of the AEB system parameters.

Reference VEB low VEB mid VEB high Perfect road Early activation

TTClimit (s) 1 1 1 1 1 2
w (m) 1 1 1 1 1 2
amax (m/s2) g*μ 1.3*g*μ 1.55*g*μ 1.8*g*μ 0.9*g g*μ

316
H. Jeppsson et al. Accident Analysis and Prevention 111 (2018) 311–320

Table 4
System effectiveness (%) of fatal, MAIS3+F, MAIS2+F and collision avoidance.

Reference VEB low VEB mid VEB high Perfect road Early activation

Fatal 71.6 79.8 84.5 87.3 78.2 84.2

MAIS3+F 71.5 80.8 86.0 89.0 79.1 84.5
MAIS2+F 64.2 75.9 82.6 86.5 74.8 80.0
Collision avoidance 53.9 68.1 76.9 81.9 66.6 72.9

Fig. 5 plots TTC and vehicle speed at visibility with theoretical

avoidance limits at different deceleration levels. One can see that in line
with the avoidance predictions, 92% could theoretically be avoided
with a deceleration of 10 m/s2 and 97% with 20 m/s2. Few cases re-
quire more than 20 m/s2 to prevent collision.

4.3. Analysis of remaining collisions

The distribution of all accident scenarios is given in Fig. 6. Also

Fig. 5. Vehicle speed at visibility vs TTC at visibility. Blue dots represent the complete
data sample. Black lines are theoretical avoidance limits. A dot above a line means that
the collision could be avoided. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
cases with high impact speed remain. This indicates that increasing
deceleration levels further will result in only a small improvement in
effectiveness. The high effectiveness of VEB can be explained by the
early visibility of pedestrians: only a few cases of obstructed view and
late TTC at visibility were recorded in the data and hence there was
sufficient time for speed to be reduced.
included are the distributions of collisions not avoided – the remaining
collisions – for the two system settings “Reference” and “VEB high”.
Introducing “Reference” and “VEB high” does not dramatically change
the accident scenario distribution for the remaining collisions. How-
ever, accidents with vehicle turning and pedestrian walking from op-
posite direction are almost eliminated with the reference AEB setting
and completely eliminated with “VEB high”.
Cases where the pedestrian will still impact with “VEB high” are
characterised by high pedestrian walking speed and late visibility (see
Fig. 7a and b). The reason for late visibility in many cases is that pe-
destrian view was obstructed. The majority of remaining collisions at
“VEB high” are corner impacts which is a consequence of high pedes-
trian walking speed and late visibility (Fig. 7d).
5. Discussion
VEB with a maximum deceleration in the middle of the performance
range (“VEB mid”) has an effectiveness similar to that of an “Early
activation” system, where the AEB is triggered as early as 2 s TTC. This

Fig. 6. Accident scenarios for all collisions and remaining collisions for ‘reference’ and “VEB high”.

317
H. Jeppsson et al.
Fig. 7. (a) Pedestrian impact speed. (b) Time to Collision at Point of Visibility. (c) Vehicle impact speed. (d) Distribution of frontal impact for “VEB high”.
is important as earlier activation and increased false positives rates go
hand in hand (Helmer, 2014), mainly due to increased uncertainty in
vehicle and pedestrian motion until the predicted collision (Seiniger
et al., 2013). The “Early activation” system’s advantage of higher injury
prevention performance will therefore be accompanied by the dis-
advantage of higher false positive rates. VEB on the other hand offers a
substantial increase in performance without increasing false positive
rates compared to the reference system as it does not alter activation
time.
5.1. Strengths and limitations
A particular strength of the study is the number of cases it is based
on; the 526 cases utilised here by far exceeds the 100 reconstructed
cases used in simulation by Hamdane et al. (2015). There is a greater
likelihood of capturing surprise events with this amount of data (Sander
and Lubbe, 2016), and estimates of effectiveness should be more reli-
able than those based on substantially smaller sets of test scenarios, as
for example in Lubbe and Kullgren (2015).
Limitations of this study include the use of a simple vehicle dy-
namics model with ideal sensors which showed no performance de-
gradation due to weather conditions, light conditions or driving speeds.
The AEB algorithm is realistic but relatively simple. In other algorithms,
AEB activation parameters might depend on driving speeds, requiring
earlier activation at higher and later activation at lower speeds (e.g.
Hayashi et al., 2013). Other algorithms may give more consideration to
the ability of pedestrians to change their trajectory or stop suddenly,
and therefore activate AEB only when a pedestrian is assumed not to be
able to avoid a collision (e.g. Seiniger et al., 2013).
The result may be affected by the fact that the data in PCM is likely
318
Accident Analysis and Prevention 111 (2018) 311–320
to under-report view obstructions as moving traffic is not recorded. This
may result in an overestimation of the predictability of pedestrian paths
and collisions.
This study analyses potential rather than real system performance
despite our best efforts to set realistic system parameters (see Appendix
A). However, as it compares different systems to which these limita-
tions apply equally, relevant conclusions can still be drawn.
Furthermore, no investigation into the effectiveness of other technolo-
gies which aim to reduce pedestrian casualties, such as V2X commu-
nication or steer avoidance, has yet been undertaken. No quantification
of the relative effectiveness of VEB compared to competing technologies
can therefore be made.
5.2. Data quality and model fit
AEB is only effective in frontal impacts, thus we filtered the accident
data to obtain only these. Filtering the GIDAS data such that the main
deformed vehicle area (VDI2) is the front resulted in some cases where
the pedestrian walks into the side of the vehicle. The reason why this is
coded as front is probably because the pedestrian later falls onto the
bonnet or the windscreen. These cases are not of interest to this study;
the method adopted, basing the filtering criteria on the PCM data in-
stead, gives a result that fits this study better. The numbers of cases are
almost the same with the two different methods and 90% of the cases
are the same.
The model fit assessment for the injury risk curves shows that model
fit is reasonable (see Appendix C).
An alternative weighting, calculating weight factor year-by-year
instead of a weight factor for all years together, and the resulting
changes to injury risk curves, were assessed. No major differences can
H. Jeppsson et al.
be seen. A comparison between the risk curves obtained here and those
reported by Rosén (2013) was also made. No major differences were
seen for either fatal or MAIS3+F, despite the fact that Rosén filtered
the data differently, for example by excluding sport utility vehicles,
other light trucks and vans, and children under 15 years of age, and did
not restrict year of registration.
The MAIS2+F injury curve does not pass through the origin, an
often observed issue when using logistic regression for injury risk
curves. Schramm (2011) argues that logistic regression is nevertheless a
suitable functional description of the injury risk and defining a dis-
continuity at the origin would be appropriate. Substantial injury risk at
low speeds reflects a reality in accident data (Schramm, 2011) and may
originate from falling and impacting road surfaces. We follow the ar-
gument and assume no injury if the impact is avoided (zero impact
speed) and otherwise apply the risk curve.
The MAIS2+F injury risk curve is less steep compared with
MAIS3+F and fatal risk curves as indicated by Fig. 4 and the coefficient
b1 in Table 3. There could be several explanations; one hypothesis is
that MAIS2+F injury risk is less dependent on impact speed as it spans
lower impact speeds where other factors (for example ground contact)
have a larger influence on the injury outcome. Another hypothesis is
that the injury risk level is steeper at higher speeds since bottoming out
has occurred. Bottoming out describes a situation where the available
deformation space between outer (softer) structures and harder un-
derlying parts of a vehicle has been used up and contact with under-
lying hard structures is made. In these situations, injury risk increases
rapidly (Hutchinson et al., 2012). The higher the impact speed, the
more deformation space is needed, and the more likely bottoming out
becomes. As Euro NCAP tests at 40 km/h, bottoming out can be ex-
pected to occur above but close to 40 km/h.
5.3. Considerations for implementing VEB in new vehicles
Several areas need careful consideration when implementing a new
protection system such as VEB. The extra weight, the packaging space
required, and positioning to ensure no negative interaction with vehicle
dynamics, are three of the greatest challenges. The performance of the
VEB is dependent on the effective area. In this study three different
dimensions were evaluated: 0.4 × 0.4 m, 0.5 × 0.5 m and 0.6 × 0.6 m.
Fitting the large version may be problematic, but already the smallest
shows good effectiveness. To transfer the extra braking force from the
VEB, stiff mountings to the vehicle structure are needed. It is also im-
portant to consider the position under the vehicle regarding exposure to
water and dirt for an optimal installation.
False positive VEB activation may occur. To limit the number of
false activations, a conservative trigger logic should be used. As an
example, if a collision can be avoided with the AEB system without the
VEB, it should not be activated, even if activation of the VEB would
mean that the brake decision could be taken later. Instead the “normal”
AEB trigger logic should be used and if, for example, friction is less than
expected and a collision is still immanent, the VEB should then be ac-
tivated. Special cases where the VEB could have a major effect on the
end result are cases with late visibility, i.e. obscured situations and
cases with very short exposure such as crossing scenarios.
5.4. Implications
Equipping all M1 vehicles with “VEB high”, VEB with the maximum
deceleration assessed here, would prevent 82% of pedestrian frontal
accidents, as shown in Table 4. Frontal accidents with M1 vehicles re-
present 48% of all pedestrian accidents in GIDAS. Therefore, the col-
lision avoidance effectiveness of “VEB high” can be calculated as
48%*82% = 38%.
Similarly, the effectiveness of “VEB high” for all pedestrian fatalities
can be calculated as the product of its effectiveness in preventing fatal
pedestrian accidents and the proportion of M1 vehicle involvement in
319
Accident Analysis and Prevention 111 (2018) 311–320
such accidents, i.e., 56%*87%, giving 49%. Hence, one can predict the
number of pedestrian fatalities in Germany in 2015 that would have
been avoided, had all M1 vehicles been thus equipped, as
537*49% = 263. This is 46 more fatalities prevented than the
56%*72%*537 = 217 cases preventable with the “Reference” AEB-only
system.
This can be applied to global accident statistics. Assuming the same
ratio of M1 involvement and the same potential “VEB high” effective-
ness globally, about 24,000 more fatalities could be avoided by using
the “VEB high” compared to the “Reference” AEB system annually
(270,000*56%*(87.3%-71.6%) = 24,000).
Euro NCAP has been rating AEB VRU protection in cars since 2016
(Schram et al., 2015). The test scenario consists of the vehicle traveling
straight and a pedestrian crossing from the left or the right, with or
without obstruction. The sample data can be filtered for this accident
scenario which results in a total of 426 accidents. Of these, 47% would
be avoided with the reference AEB system and 80% would be avoided
with the “VEB high” system (see Fig. 6). Discussions of rating other AEB
VRU scenarios, such as those in which the vehicle is turning and the
pedestrian is crossing, is ongoing, for example in the Prospect project
(Seiniger et al., 2016). Filtering the sample data for situations with a
turning vehicle and a pedestrian crossing in the same direction results
in a total of 41 accidents, of which 48% could be avoided with the
reference system installed, and 78% avoided with the “VEB high”
system. For scenarios where the vehicle is turning and the pedestrian is
crossing in the opposite direction, all cases would be prevented with
VEB. It should be possible to assess the benefit of VEB in all current or
future scenarios used in Euro NCAP testing.
5.5. Future work
Even though the effectiveness of the VEB is high, 27% of pedestrian
to M1 accidents are cases where the pedestrian is not impacted by the
front of the vehicle, and which therefore cannot be addressed by any
AEB system that only brakes for pedestrians projected to impact the
front of the car. There is a clear need, therefore, also to address colli-
sions where the pedestrian is impacted by the side of a car, and by a
reversing car.
The VEB functionality has great potential to increase the effective-
ness of systems other than pedestrian-focused AEB systems, such as
rear-end AEB systems and those developed for crossing vehicle-to-ve-
hicle scenarios, or for vehicles colliding with cyclists. VEB is likely to be
especially effective in scenarios when the time to activation is short, i.e.
crossing and obstructed accidents, or when the friction is low; the ef-
fectiveness should be quantified in future work. Another application
area that also might be of interest is trucks colliding with the end of a
traffic jam.
6. Conclusion
This study has introduced and modelled a braking system which
increases deceleration up to a maximum of 16 m/s2 using PCM accident
data. Pedestrian AEB with VEB decreases pedestrian fatalities in colli-
sions with M1 vehicle fronts by 80–87%, which is an increase of at least
8% over the AEB-only system. Had all M1 vehicles been equipped with
AEB and a high performance VEB, approximately 215 pedestrian
fatalities could have been avoided in Germany in 2015, 46 more than
with AEB only. Collision avoidance increases by 14–28% when VEB is
added to the reference AEB system.
Most collisions and injuries can be avoided when AEB is supple-
mented by the high performance VEB; remaining cases are char-
acterised by high pedestrian walking speeds and late visibility due to
view obstructions. VEB is effective in all analysed accident scenarios,
and we believe that the real-life performance increase of VEB over AEB
that we have demonstrated here would also be demonstrated in Euro
NCAP pedestrian AEB tests.
H. Jeppsson et al.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in the
online version, at https://doi.org/10.1016/j.aap.2017.12.001.
References
Bärgman, J., Boda, C.N., Dozza, M., 2017. Counterfactual simulations applied to SHRP2
crashes: the effect of driver behavior models on safety benefit estimations of in-
telligent safety systems. Accid. Anal. Prev. 102, 165–180. http://dx.doi.org/10.1016/
j.aap.2017.03.003.
Bärgman, J., 2015. On the Analysis of Naturalistic Driving Data. Chalmers university of
technology.
Bärgman, J., 2016. Methods for Analysis of Naturalistic Driving Data in Driver Behavior
Research. Chalmers university of technology.
Bhalla, K., Shotten, M., Cohen, A., Brauer, M., Shahraz, S., Burnett, R., Leach-Kemon, K.,
Freedman, G., Murray, C.J.L., Dingenen, R., Van Dentener, F., Vos, T., Naghavi, M.,
Abraham, J., Bartels, D., Weh, P.-H., 2014. Transport for Health: The Global Burden
of Disease from Motorized Road Transport.
Bundesministerium für Verkehr und digitale Infrastruktur, 2015. Halbzeitbilanz des
Verkehrssicherheitsprogramms. pp. 2011–2020.
Bundestag, Deutscher, 2016. Deutscher Bundestag Unterrichtung Drucksache 18/9640.
Cicchino, J.B., 2017. Effectiveness of forward collision warning and autonomous emer-
gency braking systems in reducing front-to-rear crash rates. Accid. Anal. Prev. 99,
142–152. http://dx.doi.org/10.1016/j.aap.2016.11.009.
Coles, S., 2001. An Introduction to Statistical Modeling of Extreme Values. Springer,
London, United Kingdom).
Edwards, M., Nathanson, A., Wisch, M., 2014. Estimate of potential benefit for Europe of
fitting Autonomous Emergency Braking (AEB) systems for pedestrian protection to
passenger cars. Traffic Inj. Prev. 15 (Suppl. 1), S173–82. http://dx.doi.org/10.1080/
15389588.2014.931579.
Edwards, M., Nathanson, A., Carroll, J., Wisch, M., Zander, O., Lubbe, N., 2015.
Assessment of integrated pedestrian protection systems with autonomous emergency
braking (AEB) and passive safety components. Traffic Inj. Prev 16 (Suppl. 1), S2–S11.
http://dx.doi.org/10.1080/15389588.2014.1003154.
Erbsmehl, C., 2009. Simulation of real crashes as a method for estimating the potential
benefits of advanced safety technologies. In: The 21 St International Technical
Conference on the Enhaced Safety of Vehicles (ESV) Stuttgart. Germany.
European Commission, 2003. Directive 2003/102/ec.
European Commission, 2010. Towards a European Road Safety Area: Policy Orientations
on Road Safety 2011–2020. COM 389 Final.
European Commission, 2015. Interim Evaluation of the Policy Orientations on Road
Safety. pp. 2011–2020.
European Commission, 2016. Traffic Safety Basic Facts 2016 Pedestrians. http://dx.doi.
org/10.1136/bmj.330.7487.367.
European Enhanced Vehicle-Safety Committee, 2002. EEVC Working Group 17 Report.
Improved Test Methods to Evaluate Pedestrian Protection Afforded by Passenger
Cars.
Farah, H., Azevedo, C.L., 2017. Safety analysis of passing maneuvers using extreme value
theory. IATSS Res. 41 (1), 12–21. http://dx.doi.org/10.1016/j.iatssr.2016.07.001.
Fildes, B., Keall, M., Bos, N., Lie, A., Page, Y., Pastor, C., Pennisi, L., Rizzi, M., Thomas, P.,
Tingvall, C., 2015. Effectiveness of low speed autonomous emergency braking in real-
world rear-end crashes. Accid. Anal. Prev. 81, 24–29. http://dx.doi.org/10.1016/j.
aap.2015.03.029.
Fredriksson, R., Rosén, E., 2012. Integrated pedestrian countermeasures – potential of
head injury reduction combining passive and active countermeasures. Saf. Sci. 50 (3),
400–407. http://dx.doi.org/10.1016/j.ssci.2011.09.019.
Hamdane, H., Serre, T., Masson, C., Anderson, R., 2015. Issues and challenges for pe-
destrian active safety systems based on real world accidents. Accid. Anal. Prev. 82,
53–60. http://dx.doi.org/10.1016/j.aap.2015.05.014.
Hayashi, H., Inomata, R., Fujishiro, R., Ouchi, Y., Suzuki, K., Nanami, T., 2013.
Development of pre-Crash safety system with pedestrian collision avoidance assist.
In: The 23rd International Enhanced Safety of Vehicles (ESV) Conference. Seoul,
Republic of Korea. pp. 8.
Helmer, T., 2014. Development of a Methodology for the Evaluation of Active Safety
Using the Example of Preventive Pedestrian Protection. Technische Universität,
Berlin.
Hutchinson, T.P., Anderson, R.W.G., Searson, D.J., 2012. Pedestrian headform testing:
inferring performance at impact speeds and for headform masses not tested, and
estimating average performance in a range of real-world conditions. Traffic Inj. Prev.
13 (4), 402–411. http://dx.doi.org/10.1080/15389588.2012.660252.
Hydén, C., 1987. The Development of a Method for Traffic Safety Evaluation: The
Swedish Traffic Conflicts Technique. Lund University, Department of Traffic Planning
and Engineering, Lund, Sweden).
Jonasson, J.K., Rootzén, H., 2014. Internal validation of near-crashes in naturalistic
driving studies: a continuous and multivariate approach. Accid. Anal. Prev. 62,
102–109. http://dx.doi.org/10.1016/j.aap.2013.09.013.
Kompass, K., 2012. New methods for representative evaluation of integral safety in traffic.
In: 5th International Conference on ESAR, Expert Symposium on Accident Research.
Hannover, Germany. pp. 21.
320
Accident Analysis and Prevention 111 (2018) 311–320
Korner, J., 1989. A Method for Evaluating Occupant Protection by Correlating Accident
Data with Laboratory Test Data. SAE Tech. (Pap. 890747).
Kullgren, A., 2008. Dose-response models and EDR data for assessment of injury risk and
effectiveness of safety systems. In: Proceedings of IRCOBI Conference. Bern,
Switzerland. pp. 3–14.
Kusano, K.D., Gabler, H.C., 2012. Safety benefits of forward collision warning, brake
assist, and autonomous braking systems in rear-end collisions. IEEE Trans. Intell.
Transp. Syst. 13 (4), 1546–1555. http://dx.doi.org/10.1109/TITS.2012.2191542.
Lang, J., 2015. Vacuum emergency brake introduction: the torricelli brake. In: The 1 st
Praxis Conference Autonomous Emergency Braking. Bicester, United Kingdom.
Lindman, M., Oedblom, A., Bergvall, E., Eidehall, A., Svanberg, B., Lukaszewicz, T., 2010.
Benefit estimation model for pedestrian auto brake functionality. In: The 4th
International Conference Expert Symposium on Accident Research (ESAR).
Hannover, Germany. pp. 28–33.
Lubbe, N., Kullgren, A., 2015. Assessment of integrated pedestrian protection systems
with forward collision warning and automated emergency braking. In: Proceedings of
IRCOBI Conference. Lyon, France. pp. 385–397.
Mellinghoff, U., Breitling, P.D.T., Schöneburg, P.D.R., Metzler, H.-G., 2009. 2009. the
Mercedes-Benz experimental safety vehicle ESF. In: The 21 st International Technical
Conference on the Enhaced Safety of Vehicles (ESV). Stuttgart, Germany.
Morales Teraoka, E.Y., Tanaka, S., 2014. Benefit estimation of active safety systems for
crossing-Pedestrian scenarios. In: Proceeding of FISITA 2014 World Automotive
Congress. Maastricht, Netherlands.
Morales Teraoka, E.Y., Tanaka, S., Mochida, T., 2013. Benefit estimation of active safety
systems for crossing-Pedestrian scenarios. In: Proceedings of the 1 st International
Symposium on Future Active Safety Technology Towards Zero Traffic Accidents.
Nagoya, Japan. pp. 1–7.
Ohlin, M., Strandroth, J., Tingvall, C., 2017. The combined effect of vehicle frontal de-
sign, speed reduction, autonomous emergency braking and helmet use in reducing
real life bicycle injuries. Saf. Sci. 92, 338–344. http://dx.doi.org/10.1016/j.ssci.
2016.05.007.
Pastor, C., 2013. The correlation between pedestrian injury severity in real-life crashes
and Euro NCAP pedestrian test results. Traffic Inj. Prev. 12 (6), 13–0308. http://dx.
doi.org/10.1080/15389588.2011.607198.
Rosén, E., Källhammer, J.E., Eriksson, D., Nentwich, M., Fredriksson, R., Smith, K., 2010.
Pedestrian injury mitigation by autonomous braking. Accid. Anal. Prev. 42 (6),
1949–1957. http://dx.doi.org/10.1016/j.aap.2010.05.018.
Rosén, E., 2013. Autonomous emergency braking for vulnerable road users. In:
Proceedings of IRCOBI Conference. Gothenburg, Sweden. pp. 618–627.
Roth, F., Schneider, A., Labenski, V., Ag, A., Bejan, P., Coulongeat, F., Clement, B., 2017.
General safety Regulation–Future requirements in VRU safety. In: 12th Praxis
Conference Pedestrian Protection. Berglisch-Gladbach, Germany.
Sander, U., Lubbe, N., 2016. Prediction of accident evolution by diversification of influ-
ence factors in computer simulation: opportunities for driver warnings in intersection
accidents. In: Aktive Sicherheit Und Automatisiertes Fahren ? Methodenentwicklung
Im Expertendialog. Essen, Germany. pp. 29.
Sander, U., 2017. Opportunities and limitations for intersection collision intervention—A
study of real world ‘left turn across path’ accidents. Accid. Anal. Prev. 99, 342–355.
http://dx.doi.org/10.1016/j.aap.2016.12.011.
Schram, Richard, Williams, A., van, R.M., 2015. Euro ncap’s first step to assess autono-
mous emergency braking (aeb) for vulnerable road users. In: 24th International
Enhanced Safety of Vehicles (ESV) Conference. Gothenburg, Sweden. pp. 1–7.
Schramm, S., 2011. Methode Zur Berechnung Der Feldeffektivität Integraler
Fußgängerschutzsysteme.
Seiniger, P., Bartels, O., Pastor, C., Wisch, M., 2013. An open simulation approach to
identify chances and limitations for vulnerable road user (VRU) active safety. Traffic
Inj. Prev. 14, S2–12. http://dx.doi.org/10.1080/15389588.2013.797574.
Seiniger, P., Bartels, O., Hellmann, A., Fritz, M., 2016. Test Protocol as a Proposal for
Consumer Testing.Test Protocol as a Proposal for Consumer Testing.
Strandroth, J., Rizzi, M., Sternlund, S., Lie, A., Tingvall, C., 2011. The correlation between
pedestrian injury severity in real life crashes and Euro NCAP pedestrian test results.
Traffic Inj. Prev. 12 (6), 604–613.
Tanaka, S., 2015. Reconstructing accidents by simulation for developing active safety
systems. In: Proceedings of the 3rd International Symposium on Future Active Safety
Technology Towards Zero Traffic Accidents. Gothenburg, Sweden. pp. 597–603.
Tarko, A.P., 2012. Use of crash surrogates and exceedance statistics to estimate road
safety. Accid. Anal. Prev. 45, 230–240. http://dx.doi.org/10.1016/j.aap.2011.07.
008.
Wille, J.M., Zatloukal, M., 2012. rateEFFECT: effectiveness evaluation of active safety
systems. In: 5th International Conference on ESAR, Expert Symposium on Accident
Research. Fachverlag NW in Der Carl Schuenemann Verlag GmbH ISSN . 0943-9293.
Hanover, Germany. pp. 1–41.
World Health Organisation, 2015. Global Status Report on Road Safety 2015, WHO
Library Cataloguing-in-Publication Data Global. http://dx.doi.org/10.1136/
injuryprev-2013-040775.
Yanagisawa, M., Swanson, E., Azeredo, P., Najm, W.G., 2017. Estimation of Potential
Safety Benefits for Pedestrian Crash Avoidance/Mitigation Systems. National
Highway Traffic Safety Administration report, Washington, DC).
van Ratingen, M., Williams, A., Lie, A., Seeck, A., Castaing, P., Kolke, R., Adriaenssens, G.,
Miller, A., 2016. The european new car assessment programme: a historical review.
Chin. J. Traumatol. (English Ed.) 19 (2), 63–69. http://dx.doi.org/10.1016/j.cjtee.
2015.11.016.