Sie sind auf Seite 1von 12

Transportation Research Part F 43 (2016) 2435

Contents lists available at ScienceDirect

Transportation Research Part F


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

How drivers fail to avoid crashes: A risk-homeostasis/


perception-response (RH/PR) framework evidenced by visual
perception, electrodermal activity and behavioral responses
Yutao Ba a,b,, Wei Zhang b, Alan H.S. Chan c, Tingru Zhang c, Andy S.K. Cheng d
a
IBM Research, China
b
State Key Laboratory of Automobile Safety and Energy, Department of Industrial Engineering, Tsinghua University, Beijing, China
c
Department of Systems Engineering and Engineering Management, City University of Hong Kong, Hong Kong
d
Ergonomics and Human Performance Laboratory, Department of Rehabilitation Sciences, Hong Kong Polytechnic University, Hong Kong

a r t i c l e i n f o a b s t r a c t

Article history: Drivers perception-cognition-response mechanism before the hazards developing into
Received 2 November 2015 actual crashes is a key issue of driving safety. Previous explanations on this topic are gen-
Received in revised form 21 August 2016 erally based on the perception-response (PR) process or risk-homeostasis (RH) mechanism.
Accepted 22 September 2016
However, a micro-description and related evidences as the bridge between the existing
Available online 6 October 2016
theoretical frameworks are limited. In order to generate a better understanding of crash
development, the present study examined the visual perception, electrodermal activity
Keywords:
and behavioral responses in a uniformed hazardous situation during simulator driving.
Risk homeostasis
Behavioral adaptation
Ninety-seven drivers were recruited to complete a pre-defined driving task containing a
Visual detection baseline scenario and a paired hazard scenario. The results demonstrated that the crash
Anticipatory hazard and no-crash drivers showed similar performances in terms of PR time. However, two dri-
Perceived risk ver groups showed differed hazard anticipation (indicated by aroused electrodermal activ-
Crash ity) and demonstrated distinguishable driving behaviors (measured by gas pedal, velocity
and time-to-collision) before the hazard was visually detected. These findings evidenced
the anticipatory risk perception and the subsequent concurrent functions of the RH mech-
anism and PR process at the critical moment before the occurrence of possible crashes.
Therefore, a fused RH/PR framework was proposed to describe the drivers control loop
during crash avoidance. Accordingly, several interventions were suggested to reduce crash
risk.
2016 Elsevier Ltd. All rights reserved.

1. Introduction

The injuries and deaths as a result of traffic accidents are serious social issues with the whole world concerned about.
Each year, more than one million lives are terminated by the traffic crashes worldwide (WHO, 2013). Specifically, the drivers
responses in the transient moment under the potential hazards are vital to determine whether a driver will be involved in a
crash or not. In order to eliminate the on-road tragedies, researchers have conducted significant number of studies to
expound the drivers perception-cognition-response process during such critical moment (Fuller, 2005; Kinnear, Kelly,

Corresponding author at: IBM Research, China.


E-mail address: bytbabyt@cn.ibm.com (Y. Ba).

http://dx.doi.org/10.1016/j.trf.2016.09.025
1369-8478/ 2016 Elsevier Ltd. All rights reserved.
Y. Ba et al. / Transportation Research Part F 43 (2016) 2435 25

Stradling, & Thomson, 2013; Lee, 2009; Lemercier et al., 2014; Olson & Sivak, 1986; Shahar, Poulter, Clarke, & Crundall, 2010;
Taylor, 1964; Volkow & Baler, 2012).

1.1. Perception-response process

The most direct and instinctive explanations for the failure of crash avoidance is the delaying or impaired behavioral
responses to approaching hazards, which were detailed under the framework of perception-response (PR) process (Olson
& Sivak, 1986). The PR time refers to the period from a potential hazard first becomes visible until the driver initiates the
brake reaction, which is the key component of safety margin or stop distance (Dozza, 2013; Kiefer, LeBlanc, & Flannagan,
2005; Olson & Sivak, 1986; Summala, Lamble, & Laakso, 1998).
A number of studies focused on the detection stage (i.e., from the initiation of visible hazard to the point that the hazard is
visually detected) have the general findings that drivers deficient perception within the PR process is related to their inex-
perience (Borowsky, Shinar, & Oron-Gilad, 2010; Liu, Hosking, & Lenn, 2009; Scialfa et al., 2011) or visual/mental distraction
(Haque & Washington, 2014; Muhrer & Vollrath, 2011). For instance, it was shown that experienced drivers demonstrated
significantly shorter time to detect a hidden hazard than novice drivers, which protects them against abrupt hazards
(Crundall, Underwood, & Chapman, 1999; McKnight & McKnight, 2003); the non-driving task information processing (e.g.,
mobile phone or on-board information system) would occupy the drivers visual/cognitive resource, and in turn weaken their
perceptive awareness of on-road hazards (Martens & Brouwer, 2013; Muhrer & Vollrath, 2011). Recent empirical evidences
further suggested that the time interval of reaction stage (i.e., from the time that the hazard is detected to the start point of
behavioral responses) is also affected by driving experience (Huestegge, Skottke, Anders, Msseler, & Debus, 2010), as well as
other visual/cognitive tasks (DAddario, Donmez, & Ising, 2014). In general, these studies within the scope of PR process sug-
gested the longer response time, either in the detection stage or reaction stage, provided reasonable explanations for the
driver populations with higher accident rate.

1.2. Risk-homeostasis and behavioral adaptation

The PR process assumes that all individuals will adopt the identical response strategy after visually perceiving the unex-
pected stimuli of hazards. However, this assumption might be violated as the hazards in real traffic situations are much more
complex and dynamic (Fuller, 2005). The early study of Taylor (1964) found that as indicated by the electrodermal activity,
the drivers would continuously adjust their driving behaviors (e.g., velocity) according to their subjective feeling and apprai-
sal. Such subjective feeling refers to the drivers own estimate of the probability that the potential hazards would develop
into a crash. In order to maintain the subjective risk estimate of dynamic traffic situation under a certain threshold, drivers
adjust driving behaviors accordingly, which is interpreted as behavioral adaptation according to the risk homeostasis (RH)
theory (Fuller, 2005; Fuller, 2011; Trimpop, 1996; Wilde, 1989).
Such generalized framework offers a reasonable description that the subjective appraisal of drivers on current driving sit-
uation plays an important role in the perception-cognition-response loop to avoid themselves from fatal risks (Bechara,
Damasio, Tranel, & Damasio, 1997; Volkow & Baler, 2012). In this loop, risk compensation or behavioral adaptation based
on the perceived risk level deceases the objective probability of crashes when drivers encounter the potential hazards.
For instance, when a driver is traveling in the complex traffic situation (e.g., high frequent conflicts with others) or conduct-
ing dual tasks during driving (e.g., mobile phone conversation), the slow velocity and conservative manners are often
observed (Fuller, 2005); when the vehicle is traveling on the highway with no potential obstacle and few hazard, the driver
would like to speed up (Schmidt-Daffy, 2014). In different situations, one drivers behavioral patterns are distinguishable.
However, he/she would maintain relatively consistent level of perceived risk through adopting different behavioral strate-
gies to adapt to these varied situations. Thus, if the driver underestimates the actual risks, inappropriate driving behaviors
are often adopted with high probability of crashes.

1.3. The present study

As illustrated in Table 1, either the PR process or RH framework, from different perspectives, suggested the major contrib-
utors to traffic crashes from human aspects initiated with potential hazards. The PR process provided the description with

Table 1
Contributors of crash development in sequential segments.

Segments/framework PR RH
No visible hazard Not covered Underestimated subjective appraisal of actual
Initiation of visible hazard Longer detection time/delayed responses risk/insufficient risk compensation
Visually detected hazard
Initiation of behavioral response
After behavioral response Not covered
26 Y. Ba et al. / Transportation Research Part F 43 (2016) 2435

high temporal precision to cover the moment from visible hazards to behavioral response. However, the periods before and
after were not explicitly explained, which would also relate to crash probability. On the other hand, the RH framework pro-
vided more generalized explanations of psychological mechanism to cover all stages of crash development. A recent study
with film clip testing found that the experienced drivers demonstrated more notable electrodermal activity than the novice
drivers when perceiving the developing on-road hazards (Kinnear et al., 2013), which offered a new explanation for the acci-
dent involvements of novice drivers. In line with this result, a simulator driving studies conducted by Kbler et al. (2014)
further suggested that the subjective risk estimate indicted by skin conductance level would be a better predictor of hazard
perception rather than the visual measures, which implied that the timely visual detection was not necessarily the precon-
dition of the successful crashes avoidance. These studies implied that the RH mechanism, rather than the PR process might
be functional before the crash occurrence. Inspired by above findings, a more precise description and related evidences for
the concurrent functions of two parallel mechanisms need to be further investigated, which should cover all stages of crash
development.
The primary aim of the present study is to analyze the drivers perception-cognition-response mechanism while coping
with potential hazards, by examining the synchronized measures of visual attention, electrodermal activity, and behavioral
responses in the different segments of crash development. We assume that the drivers who are involved in the crashes (i.e.
crash drivers) would statistically differ in these measures from the drivers who can successfully avoid the crashes (i.e. no-
crash drivers). To further explain the findings in such hazardous situation, the analysis of paired non-hazard driving scenario
was provided as the baseline. Finally, in order to provide a better account of the contributors for the crash involvements, we
seek to interpret these differences on the basis of the PR process and RH framework, which have been well discussed and
accepted in previous studies.

2. Methods

2.1. Participants

Ninety-seven drivers were enrolled in this study as participants via the contact list of Mercedes-Benz Academy & Feng-
shun Driving School in Beijing. Each participant received an instruction about the experimental procedure and simulator
driving task, and signed the informed consent. All participants reserved the option of withdrawing from the experiment
at any time.
During the experiment, four participants failed to finish the driving task due to the self-reported simulator sickness. Three
participants were excluded due to the failure of eye tracking data or unstable electrodermal recording. Thus, the final sam-
ples included 90 participants (68 male and 22 female), ranging in age from 18 to 55 years (Mean = 33.1, SD = 12.1). Their
averaged driving experience is 5.1 years (SD = 4.7) and 5819.1 km driving distance per year (SD = 4143.6).

2.2. Simulator driving task and scenario design

The goal of the simulator driving task for the participants was to get to the destination as soon as possible by traveling
along a designated route (approximately 14 km) in a virtual urban environment. During simulator driving, participants were

Fig. 1. Schematic representation of baseline and hazard scenarios. Projected scenarios and fixation position are illustrated in the bottom box.
Y. Ba et al. / Transportation Research Part F 43 (2016) 2435 27

required to avoid any crash and violation (e.g., speeding, running a red light and driving in illegal lane) as it required during
their daily driving. The speed limit is 60 km/h. Two specified scenarios, a baseline scenario and a hazard scenario (100 m
length for each), were located in the route at the positions of 10 km and 12 km respectively, to provide the naturalistic
but precisely controlled situations for the current analysis (illustrated in Fig. 1).
In the baseline scenario, the participant was driving along a straight path and approaching an intersection, with the line of
sight in front being unblocked. No other vehicle (named as hazard vehicle) would rush out from the intersected lane and
conflict with the participants path. Baseline scenario served as the situation with no potential hazard.
The road layout and traffic condition in the hazard scenario was identical to that in the baseline scenario, except that a
building blocked the participants front visual field and a hazard vehicle behind the building would rush out along the inter-
sected lane. When this hazard vehicle appeared in visual field, there was only 50 m distance left for the participants to have
any reaction to avoid the crashes. Hazard scenario served as a time window of data recording to describe drivers perception-
cognition-response process in the critical moment before crashes.
Referring to the PR process (Jurecki & Stanczyk, 2014), the hazard scenario was divided into four segments: (1) from the
moment that the participants vehicle entered the scenario to the moment that the hazard vehicle appeared in the visual
field, (2) form the end of segment 1 to the moment that the participant visually detected (i.e., first fixation) the hazard vehi-
cle, (3) from the end of segment 2 to the moment that the participant initiated the brake reaction, (4) from the end of seg-
ment 3 to the moment that the participant finally approached the conflict point which ended with a crash or a successful
avoidance. The distance and time period, for each segment differed between individuals according to their actual perfor-
mance in the hazard scenario.

2.3. Apparatus and measurements

2.3.1. Driving simulator and behavioral data


The portable desk-top driving simulator and measurement instruments for this study are illustrated in Fig. 2. The images
of virtual scenarios were generated by Vega Prime (Presagis Inc., Canada), subtending 100 in the horizontal direction and
75 in the vertical direction. The generated driving scenarios were projected at a resolution of 1600  1200 pixels, on the
concrete walls 150 cm ahead from the participants seat position. The simulator vehicle was operated via G25 steering wheel
and pedals (Logitech Inc., U.S.). The velocity, displacement of gas pedal, displacement of brake pedal, and time-to-collision
(TTC) to the approaching hazard vehicle were recoded continuously.

2.3.2. Eye movement recording


A RED contact-free eye tracker (SMI Inc., Germany) was used to measure the participants visual attention. Gaze data was
logged at 200 Hz based on the coordinate frame calibrated by a remote camera. The fixations on the projected driving sce-
nario were calculated through manufactory default setting. The angular error of eye tracking data was within one degree
during calibration (i.e., positional error was less than 16 pixels).

2.3.3. Skin conductance recording


The electrodermal activity in terms of skin conductance (SC) during simulator driving task was recorded by the SC-Flex
sensor of ProComp bio-recorder (Thought Technology Ltd., Canada). Previous studies have demonstrated that the high SC
indicated the increased psychological state of arousal motivated by the sympathetic nervous system (Bechara et al., 1997;
Kreibig, 2010). In addition to the raw SC signal, SC/SCbaseline (i.e., raw SC signal divided by the averaged SC level in the base-
line scenario) was also calculated to evaluate the change rate of electrodermal responses in hazard scenario and exclude the

Fig. 2. Driving simulator and instruments to measure eye movement and electrodermal activities.
28 Y. Ba et al. / Transportation Research Part F 43 (2016) 2435

individual difference of the raw signal (Mehler, Reimer, & Coughlin, 2012). Two electrode snaps were fastened around the
fingertips of index finger and little finger of the left hand, to measure the skins ability to conduct electricity. The dominant
hand of all participants was right hand to avoid the possible bias of operation habit.

2.4. Experiment procedure

Upon arrival, the participants were instructed to complete a training session (approximately 10 min) to familiarize with
the simulator operation on the virtual unban road. During the formal driving task, the only instruction was please drive along
the designated route and get to the destination as soon as possible. Meanwhile, please comply with the traffic rule and avoid any
crash. The speed limit is 60 km/h. The participants were not informed of any information about the potential hazard on road.

2.5. Data alignment and analysis

Collected datasets of driving behavior, eye tracking and electrodermal response were aligned manually through the video
recording, which provided a unique start point for the different time series data. The videos were recorded at a sampling rate
of 24 frames per second. Assuming that the error of alignment was 1 frame, the time error between synchronized datasets is
less than 0.08 s.
According to the research aim, we focused on the differences of behavioral responses visual perception and electrodermal
activity between crash drivers and no-crash drivers. Independent t tests were used to examine any significant between-
group differences in terms of the time and distance intervals across the pre-designed scenarios and segments. Besides,
the group (crash drivers and no-crash drivers, between-group)  segment (baseline scenario and segments within hazard
scenario, within-group) mixed design ANOVAs were conducted for other independent measures. Following pot-hoc test
was conducted when the significant between-group difference was established. The significance level of alpha for all analysis
in this study was 0.05.

3. Results

3.1. Behavioral responses, visual perception and skin conductance in baseline and hazard scenarios

Fig. 3 shows the raw data points of all measures against the longitudinal position in the baseline and hazard scenarios.
When the drivers were driving in baseline scenario, each measure was relatively stable. In contrast, the notable changes
of each measure were demonstrated in the hazard scenario.
When the drivers vehicle initially entered the hazard scenario, the TTC to hazard vehicle decreased sharply. Subse-
quently, drivers released the gas pedal and slowed down the velocity to moderate such decrease of TTC. After the hazard
vehicle appeared in the visual field, the driver first detected it and then started the brake reaction subsequently. Through
the whole process, the drivers raw SC fluctuated persistently. It should be noted that the behavioral adaptation was initiated
even before the hazards were visually perceived (i.e., first fixation on hazard vehicle, the second vertical line in Fig. 3).

3.2. Time and distance of each segment for crash and no-crash drivers

Among the 90 drivers, 21 of them failed to avoid the crashes in the hazard scenario, and other 69 drivers successfully
avoided collisions when the hazard vehicle rushed out. The time interval and distance of the baseline scenario and each seg-
ment of the hazard scenario are listed in Table 2.
No significant differences of the two groups were demonstrated in terms of PR time (i.e., total time of segment 2 and seg-
ment 3, crash drivers: Mean = 2.1 s, SD = 0.6 s; no-crash drivers: Mean = 3.4 s, SD = 2.4 s, t88 = 0.95, p = 0.17). After the initial
brake reaction, the time for crash drivers to approach the conflict point was significantly shorter than that for no-crash dri-
vers (i.e., time of segment 4, crash drivers: Mean = 1.1 s, SD = 0.7 s; no-crash drivers: Mean = 6.6 s, SD = 2.5 s, t88 = 9.95,
p < 0.01).
The crash drivers traveled significantly further than no-crash drivers during the PR process (i.e., total distance of segment
2 and segment 3, crash drivers: Mean = 30.6 m, SD = 8.9 m; no-crash drivers: Mean = 20.7 m, SD = 5.8 m, t88 = 5.21, p < 0.01).
The distance to conflict point after initial brake reaction was significantly shorter for the crash drivers than that of the no-
crash drivers (i.e., distance of segment 4, crash drivers: Mean = 14.9 m, SD = 9.1 m; no-crash drivers: Mean = 27.3 m,
SD = 5.1 m, t88 = 7.97, p < 0.01).

3.3. Behavioral responses, visual perception and skin conductance for crash and no-crash drivers

The ANOVA table for each measure is given in Table 3. As expected, the significant main effects of segment were demon-
strated for all measures. The significant main effect of group were only reported for velocity (F1,88 = 103.59, p < 0.01), gas
pedal displacement (F1,88 = 42.32, p < 0.01), TTC to hazard vehicle (F1,88 = 110.21, p < 0.01) and total fixation time on hazard
(F1,88 = 53.11, p < 0.01). Meanwhile, the significant group  segment interactions were established for velocity (F4,352 = 60.57,
Y. Ba et al. / Transportation Research Part F 43 (2016) 2435 29

Fig. 3. Scatter plots of (a) velocity, (b) gas pedal displacement, (c) brake pedal displacement, (d) TTC to hazard vehicle, (e) accumulated fixation time on
hazard vehicle, and (f) SC against the longitudinal position in baseline and hazard scenarios. The points collapse the raw data from all 90 participants
together. The trend curves (solid) are based on the cubic polynomial regression. The vertical lines (dash) indicate the mean positions of hazard in visual
field, first fixation on hazards and starting brake reaction, respectively.
30 Y. Ba et al. / Transportation Research Part F 43 (2016) 2435

Table 2
Time and distance of each segment for crash drivers and no-crash drivers.

Dependent variable Scenario/segment Crash driver (n = 21) No-crash driver t test


(n = 69)
Mean SD Mean SD t88 p
Time (s) Baseline 6.8 1.1 7.0 1.4 0.37 0.71
Segment 1 3.4 0.4 3.8 0.9 1.13 0.13
Segment 2 1.2 0.5 2.4 2.6 1.23 0.11
Segment 3 0.9 0.2 0.9 0.5 0.31 0.76
Segment 4 1.1 0.7 6.6 2.5 9.95 <0.01
Distance (m) Baseline 100.0 100.0
Segment 1 50.0 50.0
Segment 2 18.3 8.6 14.8 6.0 1.18 0.12
Segment 3 12.3 2.9 5.8 2.3 10.54 <0.01
Segment 4 14.9 9.1 27.3 5.1 7.97 <0.01

Table 3
ANOVA table of driving performance, fixation time on hazard and electrodermal measures.

Dependent variable Group Segment Group  segment


F df p F df p F df p
Velocity 103.59 1, 88 <0.01 655.75 4, 352 <0.01 60.57 4, 352 <0.01
Gas pedal displacement 42.32 1, 88 <0.01 112.91 4, 352 <0.01 10.96 4, 352 <0.01
Brake pedal displacementa 0.02 1, 88 0.87
TTC 110.21 1, 88 <0.01 368.72 4, 352 <0.01 28.61 4, 352 <0.01
Total fixation time on hazardb 53.11 1, 88 <0.01 342.22 1, 88 <0.01 63.48 1, 88 <0.01
SC 0.01 1, 88 0.94 31.55 4, 352 <0.01 0.34 4, 352 0.15
SC/SCbaselinec 0.26 1, 88 0.11 73.31 3, 264 <0.01 41.52 3, 264 <0.01
a
Only include segment 4.
b
Only include segment 3 and 4.
c
Not include baseline scenario.

p < 0.01), gas pedal displacement (F4,352 = 10.96, p < 0.01), TTC to hazard vehicle (F4,352 = 28.61, p < 0.01), total fixation time
on hazard (F1,88 = 63.48, p < 0.01) and SC/SCbaseline (F3,264 = 41.52, p < 0.01).
The means and error bars of each measure with significant between-group difference or group  segment interaction are
detailed from Figs. 48. When participants vehicles were approaching the hazard vehicle, TTC of both crash and no-crash
drivers significantly decreased, indicating the increased level of actual on-road risk (see Fig. 4). In comparing with the crash
drivers, the no-crash drivers conducted the behavioral interventions earlier, releasing the gas pedal (Fig. 5) and slowing
down the velocity (Fig. 6), even before the hazard vehicle appeared in visual field. These behavioral responses maintained
the no-crash drivers TTC in a higher level and protected them from the final crashes. In addition to the behavioral measures
in hazard scenario, the crash drivers also demonstrated deeper gas operation during baseline scenario (p = 0.01).
No between-group difference of total fixation time on hazard was demonstrated in segment 3 (p > 0.05, see Fig. 7). But,
the total fixation time of the crash drivers was significantly shorter than that of the no-crash drivers in segment 4 (p < 0.01).
Considering that both the time interval and distance in segment 4 for the crash drivers was significantly shorter than that of
the no-crash drivers, the safety margin was insufficient for the crash drivers to pay equivalent attention on hazard as the no-
crash drivers.
SC/SCbaseline increased continuously as approaching hazards for all drivers (Fig. 8). It should be emphasized that the nota-
ble electrodermal activity was reported for no-crash drivers just after they entered the hazard scenario (segment 1), signif-
icantly higher than that of crash drivers (p = 0.02). However, the SC/SCbaseline was significantly higher (p < 0.01) for the crash
drivers than that for the no-crash drivers until they took brake reaction and finally encountered the crashes (i.e., segment 4).
This indicated that the cognitive arousal level was raised earlier for these drivers who can avoid crashes before the actual
hazard was visually detected, which mirrored the above results of behavioral responses.

4. Discussion

This study investigated the synchronized data of visual measure, electrodermal activity and driving behavior in hazardous
situation during the simulator driving, in order to develop a comprehensive micro description of crash involvements within
the well accepted theoretical frameworks. This understanding of perception-cognition-response mechanism before crash
involvement would provide the suggestions for more effective interventions to avoid traffic crashes.
Consistent with the past studies on the PR framework (Dozza, 2013; Jurecki & Stanczyk, 2014; Kiefer et al., 2005; Olson &
Sivak, 1986), the drivers in this study demonstrated two clear stages when encountering the on-road hazard: detected the
Y. Ba et al. / Transportation Research Part F 43 (2016) 2435 31

Fig. 4. Mean and standard deviation of TTC of crash and no-crash drivers across segments. Asterisk indicates the significant between-group difference
within single segment, *p < 0.05, **p < 0.01.

Fig. 5. Mean and standard deviation of gas pedal displacement of crash and no-crash drivers across segments. Asterisk indicates the significant between-
group difference within single segment, *p < 0.05, **p < 0.01.

hazard in visual field (detection stage, named as segment 2 in hazard scenario) and conducted the brake reaction soon after-
ward (reaction stage, segment 3). However, no significant difference was reported for the time intervals of PR process
between the crash and no-crash drivers, either for the detection stage or the reaction stage. In contrast, the crash drivers
traveled significantly longer than the no-crash drivers in reaction stage, due to their higher velocity and shorter TTC before
the hazard appeared in visual field (before segment 2). Subsequently, the time and distance from the initial brake reaction to
the conflict point (segment 4) was also significantly shorter for the crash drivers, which finally caused their crashes. There-
fore, it was suggested that the driving velocity and safe margin to the potential hazards (i.e., TTC and distance) before the
hazard was firstly detected, rather than PR time, was the key issue to determine whether the driver would crash or not
in the present experiment.
This study also demonstrated that the drivers could develop early behavioral responses to potential hazards even before
the hazard could be visually detected (illustrated by the between-group difference in segment 1). When the drivers entered
the hazard scenario, some of them could realize that the visibility along their path was blocked by the building, which pos-
sibly hid certain hazards, and in turn slowed down the speed. Such behavioral adaptation based on the anticipatory risk per-
ception was formulated prior to the PR process, which gave the drivers more sufficient safe margin for afterward detection
32 Y. Ba et al. / Transportation Research Part F 43 (2016) 2435

Fig. 6. Mean and standard deviation of velocity of crash and no-crash drivers across segments. Asterisk indicates the significant between-group difference
within single segment, *p < 0.05, **p < 0.01.

Fig. 7. Mean and standard deviation of total fixation time on hazard of crash and no-crash drivers across segments. Asterisk indicates the significant
between-group difference within single segment, *p < 0.05, **p < 0.01.

and reaction. Compared with the no-crash drivers, the crash drivers failed to develop such earlier behavioral adaptation
sufficiently.
As demonstrated by the previous studies of drivers psychological responses under risk (Kinnear et al., 2013; Kbler et al.,
2014; Schmidt-Daffy, 2012; Taylor, 1964) or more general neural process of risk-aversion (Bechara et al., 1997; Stankovic,
Fairchild, Aitken, & Clark, 2014), the electrodermal response of the participants in this study was notably activated when
they perceived the current situation as high risk. It was also suggested that such increased SC level is highly correlated with
the mental state aroused by other concepts, such as high stress level (Kbler et al., 2014; Stankovic et al., 2014) or high cog-
nitive demand (Mehler et al., 2012). Back to the RH and other related theories, the drivers adjusted the driving behaviors to
maintain the subjective appraisal of their mental states in certain level or below the unaccepted threshold (Fuller, 2005;
Schmidt-Daffy, 2014; Taylor, 1964; Trimpop, 1996). Specifically, this study demonstrated that the drivers mental state
was higher aroused, either by the anticipated hazard or visually detected hazard, indicated by the increased SC level before
and after detection stage. More importantly, the no-crash drivers arousal level was raised early than that of crash drivers,
which elicited their predictive behavioral compensation to avoid the actual risk afterward.
The above findings supported that both the RH theory and PR process took functions concurrently when drivers encoun-
tered the on-road hazards, which facilitate the understanding of drivers crash involvements under a more general RH/PR
Y. Ba et al. / Transportation Research Part F 43 (2016) 2435 33

Fig. 8. Mean and standard deviation of SC/SCbaseline of crash and no-crash drivers across segments. Asterisk indicates the significant between-group
difference within single segment, *p < 0.05, **p < 0.01.

Fig. 9. RH/PR framework of drivers perception-cognition-response loop in hazardous situation.

framework (illustrated in Fig. 9). On the basis of this RH/PR framework, the subjective estimation of risk level, or perceived
risk (Bechara et al., 1997; Fuller, 2011; Taylor, 1964), could be developed from the individuals appraisals of two parallel
sources: anticipatory hazard or detected hazard. The anticipatory hazard is from the indirect hints of actual risk, which could
be indicated by the aroused electrodermal activity before the visible hazard. The detected hazard is the visually perceived
risk with high possibility to cause the crashes, which could be indicated by the aroused electrodermal activity after the vis-
ible hazard is visually perceived. The drivers behavioral adaptation was modified by the individuals perceived risk level of
current traffic situation. In order to maintain the perceived risk level and related mental arousal within the acceptable level
(Fuller, 2005; Fuller, 2011; Taylor, 1964; Trimpop, 1996), drivers would conduct varied behavioral responses accordingly: if
the perceived risk level rises slowly and slightly (e.g., entering the unfamiliar traffic situation or conducting a phone conver-
sation), the moderate responses (e.g., releasing accelerator pedal) often occurs; if the perceived risk level rises sharply (e.g., a
pedestrian rushing out suddenly), the strong responses are executed (e.g., a sudden brake). This perception-cognition-
response loop avoids the drivers from the traffic crashes caused by the potential hazards.
Therefore, the human factors of crash involvements could be explained as any failures of single or multiple components in
this loop. In this study, the crashes were mainly caused by the failures of hazard anticipation and moderate responses. As
suggested by related studies, such hazard anticipation could be correlated to the individuals driving experiences
(Crundall, Crundall, Clarke, & Shahar, 2012; Kinnear et al., 2013). However, considering other driving situations where the
hints of hazards are hidden deeply by surrounded events and no driver could develop sufficient anticipation through the
driving experiences, the timely detection and strong response would become critical for drivers to avoid such abrupt hazard.
In this situation, delaying visual perception or impaired behavioral reaction could be the primary contributors of crashes.
Based on such understanding, for more effective interventions to reduce the crash involvements from human aspect, two
approaches are recommended. One is to enhance the drivers ability to anticipate the potential hazards and the other one
is to eliminate the possible effects impairing visual detection and behavioral reaction.
34 Y. Ba et al. / Transportation Research Part F 43 (2016) 2435

5. Limitation

In this study, the electrodermal activity was adopted to indicate the drivers inherent psychological state evoked by the
perceived risk level. This was generally accepted by the previous studies of driving safety (Collet, Petit, Priez, & Dittmar,
2005; Kinnear et al., 2013; Kbler et al., 2014; Taylor, 1964), as well as the other psychological researches (Ogorevc,
Gerak, Novak, & Drnovek, 2013; Stankovic et al., 2014). However, some debates always exist for the strict notion, which
suggested other concepts, such as stress (Ogorevc et al., 2013), cognitive load (Mehler et al., 2012) or task complexity
(Fuller, 2005), to describe this aroused mental state. Due to the limitation of current psychological measures, it is still chal-
lenging to distinguish these concepts precisely (Kinnear et al., 2013). Thus, we inherit the concept of perceived risk (Wilde,
1989) in the proposed RH/PR framework.

6. Conclusion

The present study suggested that the RH theory (Wilde, 1989) and PR process (Olson and Sivak, 1986) took functions con-
currently to avoid the drivers from traffic crashes in the hazardous situations, which was evidenced by the synchronized
recording of visual perception, electrodermal activity and behavioral responses. In our experiment, the crash drivers and
no-crash drivers demonstrated the similar capacity in terms of time interval to detect the vital hazard and response accord-
ingly. However, the crash drivers did not anticipate the potential hazard sufficiently and adjust their driving behaviors
beforehand, which finally led them into crash involvements. In contrast, the no-crash drivers perceived the potential hazard
even before it actually appeared in visual field and adjusted their driving behaviors anticipatorily, which was indicated by
the higher skin conductance and lower accelerator usage. This predictive risk perception and subsequent behavioral adap-
tation provided the no-crash drivers with adequate safety margin before the actual hazard occurred. A more general logical
inference suggested that, if all drivers failed to develop the early hazard anticipation, the untimely visual detection and
behavioral response could become the main reasons of crashes. Based on these findings, we proposed a combined RH/PR
framework to describe the drivers control loop under hazards and explain the crash development from the human factors
perspective. In addition, several suggestions were offered based on this theoretical framework to develop more effective
interventions.

Acknowledgements

This study was supported jointly by the National Natural Science Foundation of China under Grant Numbers 71371103
and 31271100, and the Open Funding Project of National Key Laboratory of Human Factors Engineering, under Grant Num-
ber HF2013-K-04. The authors wish to acknowledge Qijia Peng and Musen Li for their work of data recording and alignment,
as well as the Mercedes-Benz academy & Fengshun Driving School in Beijing for their supports on the participant
recruitment.

References

Bechara, A., Damasio, H., Tranel, D., & Damasio, A. R. (1997). Deciding advantageously before knowing the advantageous strategy. Science, 275(5304),
12931295.
Borowsky, A., Shinar, D., & Oron-Gilad, T. (2010). Age, skill, and hazard perception in driving. Accident Analysis & Prevention, 42(4), 12401249. http://dx.doi.
org/10.1016/j.aap.2010.02.001.
Collet, C., Petit, C., Priez, A., & Dittmar, A. (2005). Stroop colorword test, arousal, electrodermal activity and performance in a critical driving situation.
Biological Psychology, 69(2), 195203. http://dx.doi.org/10.1016/j.biopsycho.2004.07.003.
Crundall, D., Crundall, E., Clarke, D., & Shahar, A. (2012). Why do car drivers fail to give way to motorcycles at t-junctions? Accident Analysis & Prevention, 44
(1), 8896. http://dx.doi.org/10.1016/j.aap.2010.10.017.
Crundall, D., Underwood, G., & Chapman, P. (1999). Driving experience and the functional field of view. Perception, 28(9), 10751087.
DAddario, P. M., Donmez, B., & Ising, K. W. (2014). EMG provides an earlier glimpse into the effects of cognitive distraction on brake motor response.
Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 58(1), 22002204. http://dx.doi.org/10.1177/1541931214581462.
Dozza, M. (2013). What factors influence drivers response time for evasive maneuvers in real traffic? Accident Analysis & Prevention, 58, 299308. http://dx.
doi.org/10.1016/j.aap.2012.06.003.
Fuller, R. (2005). Towards a general theory of driver behaviour. Accident Analysis & Prevention, 37(3), 461472. http://dx.doi.org/10.1016/j.aap.2004.11.003.
Fuller, R. (2011). Driver control theory: From task difficulty homeostasis to risk allostasis. In B. E. Porter (Ed.), Handbook of traffic psychology (pp. 1326). San
Diego: Academic Press.
Haque, M. M., & Washington, S. (2014). The impact of mobile phone distraction on the braking behaviour of young drivers: A hazard-based duration model.
Transportation Research Part C: Emerging Technologies, 50. http://dx.doi.org/10.1016/j.trc.2014.07.011.
Huestegge, L., Skottke, E.-M., Anders, S., Msseler, J., & Debus, G. (2010). The development of hazard perception: Dissociation of visual orientation and
hazard processing. Transportation Research Part F: Traffic Psychology and Behaviour, 13(1), 18. http://dx.doi.org/10.1016/j.trf.2009.09.005.
Jurecki, R. S., & Stanczyk, T. L. (2014). Driver reaction time to lateral entering pedestrian in a simulated crash traffic situation. Transportation Research Part F:
Traffic Psychology and Behaviour, 27(Part A), 2236. http://dx.doi.org/10.1016/j.trf.2014.08.006.
Kiefer, R. J., LeBlanc, D. J., & Flannagan, C. A. (2005). Developing an inverse time-to-collision crash alert timing approach based on drivers last-second
braking and steering judgments. Accident Analysis & Prevention, 37(2), 295303. http://dx.doi.org/10.1016/j.aap.2004.09.003.
Kinnear, N., Kelly, S. W., Stradling, S., & Thomson, J. (2013). Understanding how drivers learn to anticipate risk on the road: A laboratory experiment of
affective anticipation of road hazards. Accident Analysis & Prevention, 50, 10251033. http://dx.doi.org/10.1016/j.aap.2012.08.008.
Kreibig, S. D. (2010). Autonomic nervous system activity in emotion: A review. Biological Psychology, 84(3), 394421. http://dx.doi.org/10.1016/j.
biopsycho.2010.03.010.
Y. Ba et al. / Transportation Research Part F 43 (2016) 2435 35

Kbler, T. C., Kasneci, E., Rosenstiel, W., Schiefer, U., Nagel, K., & Papageorgiou, E. (2014). Stress-indicators and exploratory gaze for the analysis of hazard
perception in patients with visual field loss. Transportation Research Part F: Traffic Psychology and Behaviour, 24, 231243. http://dx.doi.org/10.1016/j.
trf.2014.04.016.
Lee, J. D. (2009). Can technology get your eyes back on the road? Science, 324(5925), 344346. http://dx.doi.org/10.1126/science.1168085.
Lemercier, C., Pcher, C., Berthi, G., Valry, B., Vidal, V., Paubel, P.-V., ... Maury, B. (2014). Inattention behind the wheel: How factual internal thoughts
impact attentional control while driving. Safety Science, 62, 279285. http://dx.doi.org/10.1016/j.ssci.2013.08.011.
Liu, C. C., Hosking, S. G., & Lenn, M. G. (2009). Hazard perception abilities of experienced and novice motorcyclists: An interactive simulator experiment.
Transportation Research Part F: Traffic Psychology and Behaviour, 12(4), 325334. http://dx.doi.org/10.1016/j.trf.2009.04.003.
Martens, M. H., & Brouwer, R. F. T. (2013). Measuring being lost in thought: An exploratory driving simulator study. Transportation Research Part F: Traffic
Psychology and Behaviour, 20, 1728. http://dx.doi.org/10.1016/j.trf.2013.04.002.
McKnight, A. J., & McKnight, A. S. (2003). Young novice drivers: Careless or clueless? Accident Analysis & Prevention, 35(6), 921925. http://dx.doi.org/
10.1016/S0001-4575(02)00100-8.
Mehler, B., Reimer, B., & Coughlin, J. F. (2012). Sensitivity of physiological measures for detecting systematic variations in cognitive demand from a working
memory task: An on-road study across three age groups. Human Factors, 54(3), 396412.
Muhrer, E., & Vollrath, M. (2011). The effect of visual and cognitive distraction on drivers anticipation in a simulated car following scenario. Transportation
Research Part F: Traffic Psychology and Behaviour, 14(6), 555566. http://dx.doi.org/10.1016/j.trf.2011.06.003.
Ogorevc, J., Gerak, G., Novak, D., & Drnovek, J. (2013). Metrological evaluation of skin conductance measurements. Measurement, 46(9), 29933001. http://
dx.doi.org/10.1016/j.measurement.2013.06.024.
Olson, P. L., & Sivak, M. (1986). Perception-response time to unexpected roadway hazards. Human Factors, 28(1), 9196. http://dx.doi.org/10.1177/
001872088602800110.
Schmidt-Daffy, M. (2012). Velocity versus safety: Impact of goal conflict and task difficulty on drivers behaviour, feelings of anxiety, and electrodermal
responses. Transportation Research Part F: Traffic Psychology and Behaviour, 15(3), 319332. http://dx.doi.org/10.1016/j.trf.2012.02.004.
Schmidt-Daffy, M. (2014). Prospect balancing theory: Bounded rationality of drivers speed choice. Accident Analysis & Prevention, 63, 4964. http://dx.doi.
org/10.1016/j.aap.2013.10.028.
Scialfa, C. T., Deschnes, M. C., Ference, J., Boone, J., Horswill, M. S., & Wetton, M. (2011). A hazard perception test for novice drivers. Accident Analysis &
Prevention, 43(1), 204208. http://dx.doi.org/10.1016/j.aap.2010.08.010.
Shahar, A., Poulter, D., Clarke, D., & Crundall, D. (2010). Motorcyclists and car drivers responses to hazards. Transportation Research Part F: Traffic Psychology
and Behaviour, 13(4), 243254. http://dx.doi.org/10.1016/j.trf.2010.04.008.
Stankovic, A., Fairchild, G., Aitken, M. R. F., & Clark, L. (2014). Effects of psychosocial stress on psychophysiological activity during risky decision-making in
male adolescents. International Journal of Psychophysiology, 93(1), 2229. http://dx.doi.org/10.1016/j.ijpsycho.2013.11.001.
Summala, H., Lamble, D., & Laakso, M. (1998). Driving experience and perception of the lead cars braking when looking at in-car targets. Accident Analysis &
Prevention, 30(4), 401407. http://dx.doi.org/10.1016/S0001-4575(98)00005-0.
Taylor, D. H. (1964). Drivers galvanic skin response and the risk of accident. Ergonomics, 7(4), 439451. http://dx.doi.org/10.1080/00140136408930761.
Trimpop, R. M. (1996). Risk homeostasis theory: Problems of the past and promises for the future. Safety Science, 22(13), 119130. http://dx.doi.org/
10.1016/0925-7535(96)00010-0.
Volkow, N. D., & Baler, R. D. (2012). To stop or not to stop? Science, 335(6068), 546548. http://dx.doi.org/10.1126/science.1218170.
WHO (2013). Global status report on road safety 2013: Supporting a decade of action. <http://www.who.int/violence_injury_prevention/road_safety_status/
2013/en/>.
Wilde, G. J. S. (1989). Accident countermeasures and behavioural compensation: The position of risk homeostasis theory. Journal of Occupational Accidents,
10(4), 267292. http://dx.doi.org/10.1016/0376-6349(89)90021-7.

Das könnte Ihnen auch gefallen