Paper on Hazardous Behavior

Research Article
Implementing Surrogate Safety Measures in Driving Simulator
and Evaluating the Safety Effects of Simulator-Based Training on
Risky Driving Behaviors

Eunhan Ka,1 Do-Gyeong Kim,2 Jooneui Hong,3 and Chungwon Lee 4

1Institute of Engineering Research, Seoul National University, Seoul 08826, Republic of Korea
2Department of Transportation Engineering, University of Seoul, Seoul 02504, Republic of Korea
3Korea Development Institute, Sejong 30149, Republic of Korea
4Department of Civil and Environmental Engineering, Seoul National University, Seoul 08826, Republic of Korea

Correspondence should be addressed to Chungwon Lee; [email protected]

Received 26 December 2019; Accepted 23 April 2020; Published 19 June 2020

Academic Editor: Inhi Kim

Copyright © 2020 Eunhan Ka et al. )is is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Human errors cause approximately 90 percent of traffic accidents, and drivers with risky driving behaviors are involved in about
52 percent of severe traffic crashes. Driver education using driving simulators has been used extensively to obtain a quantitative
evaluation of driving behaviors without causing drivers to be at risk for physical injuries. However, since many driver education
programs that use simulators have limits on realistic interactions with surrounding vehicles, they are limited in reducing risky
driving behaviors associated with surrounding vehicles. )is study introduces surrogate safety measures (SSMs) into simulator-
based training in order to evaluate the potential for crashes and to reduce risky driving behaviors in driving situations that include
surrounding vehicles. A preliminary experiment was conducted with 31 drivers to analyze whether the SSMs could identify risky
driving behaviors. )e results showed that 15 SSMs were statistically significant measures to capture risky driving behaviors. )is
study used simulator-based training with 21 novice drivers, 16 elderly drivers, and 21 commercial drivers to determine whether a
simulator-based training program using the SSMs is effective in reducing risky driving behaviors. )e risky driving behaviors by
novice drivers were reduced significantly with the exception of erratic lane-changing. In the case of elderly drivers, speeding was
the only risky driving behavior that was reduced; the others were not reduced because of their difficulty with manipulating the
pedals in the driving simulator and their defensive driving. Risky driving behaviors by commercial drivers were reduced overall.
)e results of this study indicated that the SSMs can be used to enhance drivers’ safety, to evaluate the safety of traffic management
strategies as well as to reduce risky driving behaviors in simulator-based training.

1. Introduction

)e worldwide number of annual fatalities in traffic crashes
reached 1.35 million each year, and this number continues to
increase steadily in the world [1]. Human errors cause about
90 percent of all road accidents [2], and the majority of
human errors involve risky driving. Drivers with risky
driving behaviors such as speeding, following other vehicles
too closely (tailgating), erratic driving, and violation of
traffic laws accounted for about 52% of severe traffic acci-
dents [3]. Moderating risky driving behaviors have been
achieved successfully using a variety of approaches that

combine education, engineering, and enforcement; this
approach to safety is known as the 3E principle [4]. Driver
education has been used extensively to reduce risky driving
behaviors. It has been reported to be an effective way to
reduce traffic accidents by detecting risky driving behaviors
and providing appropriate feedback to reduce these be-
haviors [5]. Risky driving behaviors should be measured and
evaluated quantitatively to give appropriate feedback to
drivers in order to reduce risky driving behaviors.

Current driver education programs have focused on
educating drivers about the skills and attitudes necessary to
become a safe driver. Videos and lectures about traffic

Hindawi
Journal of Advanced Transportation
Volume 2020, Article ID 7525721, 12 pages
https://doi.org/10.1155/2020/7525721

mailto:[email protected]

https://orcid.org/0000-0002-2845-5002

https://creativecommons.org/licenses/by/4.0/

https://doi.org/10.1155/2020/7525721

regulations and automobile-related knowledge, on-road
training, and simulator-based training generally have been
used in driver education programs. Videos and lectures help
drivers acquire knowledge about driving safely by providing
information about traffic regulations and the appropriate
operation of automobiles. However, these approaches to
teaching drivers have limitations in that they do not help
improve the practical skills that are required in on-road
driving [6]. On-road training with a driving instructor is an
effective method to educate drivers to drive more safely on
the road. However, even professional instruction and on-
road training cannot address all of the potential crashes of
driving because they cannot expose drivers to the various
potential collision situations that can occur on the road.

Driving simulators are used extensively as a tool to
instruct drivers to drive in a common driving environment
as well as in collision situations that would be too dan-
gerous to create in actual on-road driving [7]. )e in-
structor can design various driving scenarios including
myriads of road and traffic environments, movements of
surrounding vehicles, and collision scenarios. )erefore,
driving simulators can be used to give risky drivers re-
peated training with various collision situations. Driving
simulators can be used to measure driving behaviors
quantitatively as well as to acquire the trajectory data for
surrounding vehicles [8]. However, there are issues con-
cerning the validity of virtual simulations of real driving
environments. Current studies have shown that they have
similar patterns, but the driving behaviors in driving
simulators and on-road driving may not be the same [7, 9].
In other words, driving simulators can be useful tools for
educational purposes in driver education programs. In fact,
the driving instructors involved in a previous study thought
that one-hour simulator training was as effective as three
hours of on-road training [10].

Most research on reducing risky driving behaviors based
on driving simulators has been conducted with a focus on
drivers’ eye movements and the movements of the subject
vehicle, i.e., movements such as erratic acceleration and
deceleration, speed variation, and lane deviation [11–14].
Since risky driving behaviors cause severe road crashes, it is
necessary to evaluate the crash potential in the interactions
between the subject vehicle and surrounding vehicles, such
as following leading vehicles (car-following) and changing
lanes (lane-changing). However, few studies have evaluated
the crash potential between the subject vehicle and sur-
rounding vehicles, which would address the interactions
between vehicles [15]. )is study implemented realistic
interactions between the subject vehicle and surrounding
vehicles in a driving simulator by applying traffic flow
models to the movements of surrounding vehicles. In ad-
dition, we examined surrogate safety measures (SSMs),
which are used extensively in the field of road safety as useful
measures for assessing crash potential or severity even on
roads where no actual collisions have occurred. )e SSMs
can increase our understanding of the situations that cause
collisions. In this study, the SSMs were used to evaluate risky
driving behaviors in order to evaluate vehicles’ crash
potentials.

)e aim of this study was to determine whether the SSMs
can identify risky driving behaviors in driving simulators
and whether the SSMs are effective in improving drivers’
behaviors when the SSMs are used as evaluation measures in
the simulator-based training.

2. Review

Risky driving behaviors are defined differently by many
organizations and in many studies. Since the motivation of a
driver is difficult to determine, risky driving behaviors can
only be judged and evaluated based on the motions of ve-
hicles [16, 17]. Risky driving behaviors mean taking risks
that endanger the safety of both the driver and other road
users [18]. Generally, risky driving behaviors include
speeding, noncompliance with traffic laws, tailgating,
reckless changing speeds, erratic lane-changing, and threats
to other drivers (yelling and horn honking) [3, 16, 19].

Driving simulators have been used increasingly for
driver education because of the advantages they provide,
including the freedom to present drivers with a wide variety
of scenarios without any threat to their safety or the safety of
other people [20]. Studies on reducing or evaluating risky
driving behaviors using driving simulators have investigated
mainly risky driving behaviors in terms of the drivers’ re-
actions and the movements of vehicles. Studies of drivers’
reactions have used the movements of the eyes, the focus of
gazes, the duration of glances, and the number of fixations as
measures to evaluate drivers’ physical responses to collision
situations [11, 14, 21]. )ese studies have shown that drivers’
perceptions of conflict situations improve after they have
had driver education in which eye-tracking systems to
improve the ability of novice drivers and older drivers to
recognize situations where collisions could occur. Various
studies have used response time, pressure on the accelerator,
and pressure on the brake pedal to measure drivers’ re-
sponses to collision situations and red lights at intersections
[12, 13, 22]. Measures related to drivers’ reactions were used
mainly to evaluate risk perception rather than to reduce
risky driving behaviors.

Most of the studies related to the movements of vehicles
have focused on the movements of the subject vehicle, and
they evaluated primarily the risky behaviors of the drivers of
the subject vehicle, e.g., erratic steering control, speeding,
and tailgating. Erratic steering control involves the driver’s
sudden and unexpected changes in steering the vehicle or
how far the driver allows the vehicle to deviate from the
center of its lane. )e steering angle, steering reversal rate,
lane deviation, and mean lane position have been used in
assessing erratic steering control [13, 14, 22–24]. Velocity,
mean speed, speed variation, speeding, and acceleration are
speeding-related measures that can be used to determine the
driver’s compliance with the speed limit and the reckless
changing of speed [13, 14, 22, 25]. Evaluations of the gap
distance between vehicles and time-to-collision (TTC) with
a leading vehicle were made mainly in a car-following sit-
uation [15, 22]. However, the gap distance between vehicles
and the TTC between the subject vehicle and the leading
vehicle cannot take into consideration the accelerations and

2 Journal of Advanced Transportation

decelerations of either vehicle. Also, few studies have con-
sidered the potential for crashes between vehicles when they
are changing lanes [26].

Most studies have attempted to evaluate the effects of
simulator-based training on risky driving behaviors. )ese
studies focused principally on risky driving behaviors related
to the movements of the subject vehicle. Since most risky
driving behaviors require consideration of the subject ve-
hicle’s interactions with surrounding vehicles, it is essential
to evaluate the crash potential with one or more of the
surrounding vehicles. However, research considering the
interactions between vehicles has rarely been conducted
because the movements of surrounding vehicles would not
be implemented realistically. )is study attempted to in-
troduce the SSMs into a simulator-based training program
to evaluate the crash potential between vehicles and to re-
duce the risky driving behaviors associated with the sur-
rounding vehicles.

3. Methodology

3.1. Framework. )is study consisted of three parts: a survey
of SSMs and scenario design, a preliminary experiment, and
a simulator-based training program (see Figure 1). )e SSMs
can be classified into measures that consider only the subject
vehicle and measures that consider both the subject vehicle
and surrounding vehicles. Before the SSMs were used as
measures of driving behaviors, it was necessary to test
whether the SSMs could detect risky driving behaviors and
conservative driving behaviors in a driving simulator. )is
study conducted a preliminary experiment for the sensitivity
analysis of SSMs. )e purpose of the sensitivity analysis of
SSMs was to ensure that SSMs could detect extreme driving
behaviors, i.e., normal, conservative, and risky driving. In a
preliminary experiment, each driver was required to engage
in one of the three types of driving behaviors (normal,
conservative, and risky) in the driving simulator. Finally, this
study used a quantitative evaluation based on the SSMs to
analyze whether drivers reduced their risky driving be-
haviors after engaging in the simulator-based training
program.

3.2. Survey of Surrogate Safety Measures. Since a simulator-
based training requires immediate feedback concerning
which driving behaviors are risky in the various driving
scenarios, it is crucial to be able to calculate the SSMs used in
the simulator-based training within a short time after
driving in the driving simulators. )is study reviewed nu-
merous studies about road safety in order to investigate the
SSMs that can be used in simulator-based training, and 31
SSMs were selected as alternatives. Since 11 of the SSMs were
challenging to calculate instantaneously in driving simula-
tors or unsuitable in evaluating driving behaviors, 20 out of
the 31 SSMs were selected as implementable SSMs. For
example, the Crash Index is a measure concerning the se-
verity of a potential crash, and it is presented in the form of
the kinetic energy of the crash [27]. It is challenging to
translate kinetic energy values into an easily understandable

account of the risk associated with a given participant’s
driving behaviors. )us, this study excluded the Crash Index
from the implementable SSMs and selected implementable
SSMs as measures that can be explained easily to the drivers
in simulator-based training.

)e implementable SSMs were divided into “Relating to
the subject vehicle” and “Relating to surrounding vehicles,”
depending on whether or not the SSMs related to interac-
tions with surrounding vehicles. )e SSMs relating to the
subject vehicle can be calculated without any interactions
with surrounding vehicles (nos. 1 to 9 in Table 1). )e SSMs
relating to surrounding vehicles consider interactions with
surrounding vehicles, such as car-following situations and
lane-changing situations (nos. 10 to 20 in Table 1). )e gap
distance in the car-following situation (no. 10) was used to
confirm the validation of driving errors between the driving
simulator and on-road driving [7]. )e time-to-collision
(TTC, no. 12) was used for studies in which driving be-
haviors in critical situations were compared [15, 22, 35].
However, the gap distance and the TTC between the subject
vehicle and the leading vehicle have the limitation that the
difference between the acceleration of the subject vehicle and
the deceleration of the leading vehicle cannot be considered.
)is study used modified TTC (no. 13) and deceleration rate
to avoid crash (no. 14) to evaluate the crash potential by
considering the difference between the acceleration of the
subject vehicle and the deceleration of the leading vehicle in
situations where the subject vehicle is following the leading
vehicle. )e study in which the driving behaviors were
analyzed in the lane-changing situation used the gap dis-
tance (no. 16 and no. 17) with the surrounding vehicles in a
target lane [26]. )is study adopted the SSMs (nos.13, 14, 19,
and 20) that had not been used to reduce risky driving
behaviors in existing driving education programs to assess
the crash potential between the subject vehicle and sur-
rounding vehicles properly. )e contribution of this study is
to secure the effectiveness of driver education by capturing
interactions between a subject vehicle and surrounding
vehicles based on the simulator-based training using SSMs
and then ultimately induce the prevention and reduction of
road accidents.

)e SSMs consist of measures with a single outcome for
the dataset and measures with continuous outcomes cal-
culated at every time step of the dataset. Accumulated
speeding (AS), speed uniformity (SU), speed variation (SV),
acceleration noise (AN), and lane deviation were measured
with a single outcome and calculated after completing the
driving scenarios. Measures with continuous outcomes
should be transformed into a single representative value in
order to evaluate how risky the drivers’ driving behaviors
were.

A single representative value of SSMs can be obtained
either as the maximum (minimum) value of total outcomes,
such as Max S, i.e., the maximum velocity in a conflict
situation [33] or as the ratio of conflicts defined as exceeding
the threshold value of each measure [27, 36]. )is study
adopted the minimum value as the representative value for
SSMs related to lane-changing situations (see nos.15 to 20 in
Table 1). )e method to define the threshold value for

Journal of Advanced Transportation 3

3.2 Survey of surrogate
safety measures

Subject vehicle
Subject vehicle and
surrounding vehicles

(i)
(ii)

31 drivers (preliminary
experiment)
58 drivers (simulator-
based training)

(i)

(ii)

Specification of
driving simulator
Scenario design

(i)

(ii)

3.3 Driving simulator
and scenario design

3.4 Preliminary
experiment

Normal driving
Conservative driving
Risky driving

(i)
(ii)

(iii)

Driving before
intervention
Intervention (feedback)
Driving after
intervention

(i)

(ii)
(iii)

3.5 Simulator-based
training

3.6 Participants

Figure 1: Framework of the study.

Table 1: Description of implementable surrogate safety measures.

Relation with
surrounding vehicles

No. Surrogate safety measure Unit Description

Relating to the subject
vehicle

1 Accumulated speeding (AS) kph
)e normalized relative area (per unit length) bounded between the
speed profile values higher than the speed limit and the speed limit

line [28]

2 Speed uniformity (SU) kph
)e normalized relative area (per unit length) bounded between the

speed profile and the average speed line [28]
3 Speed variation (SV) kph )e standard deviation of the speed
4 Acceleration (%) m/s2 )e acceleration of the subject vehicle
5 Deceleration (%) m/s2 )e deceleration of the subject vehicle
6 Acceleration noise (AN) m/s2 )e root mean square deviation of the acceleration [29]
7 Lane deviation m )e standard deviation of lane position [30]
8 Yaw rate (%) °/s )e rotational velocity around the z-axis of the subject vehicle [31]
9 Lane change (%) — )e number of lane change manoeuvres completed

Relating to
surrounding vehicles

10 Gap distance (%) (GD) m
)e longitudinal distance along a travelled way between one vehicle’s

leading surface and another vehicle’s trailing surface [32]

11
Proportion of stopping
distance (%) (PSD)

—
)e ratio of the distance available for manoeuvring to that of the
necessary stopping distance to a projected point of collision [33]

12 Time-to-collision (%) (TTC) sec
)e time interval required for one vehicle to strike another object if
both objects continue on their current paths at their current speed

[32]

13 Modified TTC (%) (MTTC) sec
)e time interval required for one vehicle to strike another object if
both objects continue on their current paths at their current speed

and acceleration [32]

14
Deceleration rate to avoid

crash (%) (DRAC)
m/s2

)e deceleration required by the following vehicle to come to a timely
stop or attain a matching lead vehicle speed to avoid a rear-end crash

[34]

15 Min_Front_GD m
)e minimum value of gap distance (GD) with leading vehicle of

current lane in lane-changing situation

16 Min_Lag_GD m
)e minimum value of gap distance (GD) with lag vehicle of target

lane in lane-changing situation

17 Min_Lead_GD m
)e minimum value of gap distance (GD) with leading vehicle of

target lane in lane-changing situation

18 Min_Front_TTC sec
)e minimum value of time-to-collision (TTC) with leading vehicle

of current lane in lane-changing situation

19 Min_Lag_TTC sec
)e minimum value of time-to-collision (TTC) with lag vehicle of

target lane in lane-changing situation

20 Min_Lead_TTC sec
)e minimum value of time-to-collision (TTC) with leading vehicle

of target lane in lane-changing situation

4 Journal of Advanced Transportation

counting the number of conflicts in each of the SSMs was
derived from the existing literature [37]. )e 85th percentile
value of total participants’ driving data distribution was used
as the threshold for SSMs for which higher values indicated
more risky driving behaviors (acceleration (%), yaw rate (%),
lane change (%), gap distance (%), and proportion of
stopping distance (%)). For SSMs for which lower values
indicated more risky driving behaviors (deceleration (%),
time-to-collision (%), modified TTC (%), and deceleration
rate to avoid crash (%)), the 15th percentile value of total
participants’ driving data distribution was used as the
threshold.

3.3. Driving Simulator and Scenario Design. )e driving
simulator used in this study was mounted on a six-degree-
of-freedom motion system, with a size of
3500 × 3500 × 3500 mm. )e visual system for the driving
simulator consisted of three 43-inch full HD LED monitors,
providing a 150-degree field of view with a resolution of
5760 ×1080 pixels and a 60 Hz refresh rate. )e virtual
environment with various driving conditions was repre-
sented through the three monitors, with rear-view and side-
view mirrors visible on the center monitor and side mon-
itors, respectively (Figure 2(a)). )e vehicle dynamics were
validated based on the real motion of the Hyundai Sonata.

A part of the street grid in Seoul was implemented in a
virtual environment in order to enhance the reality of the
driving environment. )e total length of the designed route
in the scenario was 10.1 km, including freeway (2.3 km),
urban roads (6.0 km), and rural roads (1.8 km) (Figure 2(b)).
)e freeway consisted of the main freeway segment with a
posted speed limit of 110 kph and an off-ramp. )e urban
roads included ten signal intersections located every
200–400 m on a four-lane two-way road with a speed limit of
60 kph. )e rural roads were either two- or four-lane, two-
way roads with a speed limit of 80 kph.

)e movements of surrounding vehicles significantly
determine the mental load and ability to drive a vehicle. If
the movements of the surrounding vehicles were not real
enough, there is a possibility that drivers will drive a vehicle
differently than they would in actual driving, meaning that
the results and conclusions obtained from the simulation
would not be applicable in actual driving. Many studies
using driving simulators have been limited in expressing
realistic movements because the movements of the vehicles
were very strictly controlled to assess the drivers’ abilities in
certain crash situations [38]. )erefore, if the movements of
the surrounding vehicle are unrealistic and strictly con-
trolled irrespective of the movements of the subject vehicle,
it would be difficult to expect the reduction of risky driving
behaviors in actual driving on the road through simulator-
based training. In order to implement the realistic inter-
actions with surrounding vehicles, traffic flow models (i.e., a
car-following model, a lane-changing model, and a gap-
acceptance model) were modeled based on video data and
vehicle trajectory data, and then, they were applied to the
movements of the surrounding vehicles. Using the traffic
flow models had the additional benefit of showing different

movements in each trial, thereby increasing the sense of
reality and preventing participants from adapting to the
scenario [38]. )e generalized model of car-following was
estimated with data obtained from random vehicles on the
West-Hanam IC and the West-Icheon IC of the Jungbu
Highway [39]. )e lane-changing model was implemented
based on the vehicle trajectory data measured by nine video
cameras in the upper 400 m section of the Middle East IC of
the Seoul Ring Expressway for discretionary and mandatory
lane-changing. )e parameters of a logit model were esti-
mated with the gap distance and the speed of the subject
vehicle as independent variables. )e logit model was es-
timated for the gap-acceptance model at the intersection.
Data were collected using video cameras at six intersections
in Seoul to estimate the parameters of the gap-acceptance
model, and the data included the time gap, type of vehicle,
and traffic volume. )is study estimated the parameters of
the logit models for an unprotected left turn, an unprotected
right turn, and a roundabout using collected data.

3.4. Preliminary Experiment for Extreme Driving Behaviors
with SSMs. )is study conducted a preliminary experiment
to analyze the sensitivity of SSMs for extreme driving be-
haviors. Before participating in the preliminary experiment,
the participants were shown how to control a driving
simulator and performed one to three minutes of practice
driving to prevent simulator sickness and to adapt to the
virtual environment of the driving simulator. To use the
SSMs that could measure the crash potential of surrounding
vehicles in a driving simulator, the sensitivity analysis of the
SSMs was required to determine whether they could capture
risky driving behaviors. In this study, the experimental
methods that had been used in previous studies were used to
analyze the sensitivity of the SSMs to extreme driving be-
haviors. Past participants were involved in a study of ex-
treme driving behaviors that compared the difference in fuel
consumption depending on driving behaviors and com-
pared the difference in the performance of an urban network
on driving behaviors [40, 41].

Participants in the preliminary experiment were asked to
drive “normally,” “conservatively,” or “riskily.” In normal
driving, the participants drove the way they usually drive. In
the conservative driving condition, the participants were
asked to maintain a greater safe following distance, accel-
erate and decelerate as gently as possible, and keep their
speed under the speed limit. In risky driving, the participants
were required to complete their driving route within 10
minutes rather than the typical 15 minutes, to follow the
leading vehicle more closely than the recommended safe
distance, and to change lanes and the speed of the vehicle
erratically.

3.5. Simulator-Based Training to Improve Driving Behaviors.
)is study used SSMs that statistically could capture risky
and conservative driving within the simulator-based train-
ing conducted by the Korea Transportation Safety Authority
(KOTSA). )e simulator-based training consisted of three
parts, i.e., driving before the intervention, intervention

Journal of Advanced Transportation 5

(feedback based on results of driving behaviors), and driving
after the intervention.

Before the intervention, the participants drove an in-
troduction drive for 1 to 3 minutes to become accustomed to
the control of the driving simulator and the virtual envi-
ronment. Subsequently, they drove the driving scenario as
they usually would do, allowing the instructor to examine
the extent of their risky driving behaviors.

)e intervention consisted of two parts: feedback with a
video replay of the driver’s driving and a commentary video.
)e instructors provided feedback to the participants con-
cerning how risky they drove in terms of six risky driving
behaviors, i.e., speeding, reckless changing speed, rapid
acceleration and deceleration, erratic steering control, tail-
gating, and erratic lane-changing. )e speeding, reckless
changing speeds, rapid acceleration and deceleration, and
erratic steering control only assessed the movements of the
subject vehicle. In contrast, the tailgating and erratic lane-
changing assessed the crash potential with the surrounding
vehicles in the normal driving environment. In typical
drivers’ education programs, the instructor evaluates the
drivers’ driving behaviors based on the movements of the
subject vehicle and the crash potential with the surrounding
vehicles in specific situations (i.e., speeding, reckless
changing speeds, rapid acceleration and deceleration, and
erratic steering control). In the simulator-based training
using the SSMs of this study, the instructor informed the
drivers the risky driving behaviors, including situations in
which they were following a vehicle and changing lanes in a
common driving environment and were riskier than other
drivers. In other words, the contribution of this study is to
evaluate the movements of the subject vehicle as well as the
interactions between vehicles by identifying the risky driving
behaviors such as tailgating in car-following situations and
erratic lane-changing in lane-changing situations. Also, the
instructor educated the drivers about safe methods for
driving on the road to reduce the crash potential between the
vehicles. In the commentary video, videos of actual crashes

attributable to each of the six risky driving behaviors were
shown to encourage safe driving.

After the …