See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/251714333

A Collision Warning System for rear-end collision: a driving simulator study

Article in Procedia - Social and Behavioral Sciences · December 2011

DOI: 10.1016/j.sbspro.2011.08.075

CITATIONS READS

121 1,816

2 authors, including:

Francesco Bella
Università Degli Studi Roma Tre
79 PUBLICATIONS 1,919 CITATIONS

SEE PROFILE

All content following this page was uploaded by Francesco Bella on 09 March 2015.

The user has requested enhancement of the downloaded file.

 Available online at www.sciencedirect.com

Procedia Social and Behavioral Sciences 20 (2011) 676–686

14th EWGT & 26th MEC & 1st RH

A Collision Warning System for rear-end collision: a driving

simulator study
Francesco Bellaa,*, Roberta Russoa
a
Roma TRE University- via Vito Volterra n. 62 - 00146 Rome, Italy

Abstract

Collision warning and collision avoidance systems are emerging automotive safety technologies that assist drivers in avoiding
rear-end collisions. Their function is to allow the driver enough time to avoid the crash and yet avoid annoying the driver with
alerts perceived as occurring too early or unnecessary. The aim of this paper was analyzing the driver’s behavior in order to
define effective driver assistance systems which can be readily accepted by the driver. A study was performed with an interactive
fixed-base driving simulator. A sample of 32 drivers drove on a two-lane rural road. Four different driving traffic conditions were
implemented. The data recorded during the tests were analyzed to assess the safety distances required by the driver during a car-
following situation. Based on the risk perception of the driver a new collision warning algorithm was developed.
©
© 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of the Organizing Committee

Keywords:Collision Warning; Rear-end collisions; Driving simulator

1. Introduction

Road safety is certainly one topic of great interest that concerns the whole community at national, European and
global level. Several studies have been conducted to improve road safety and reduce the number of accidents on the
roads. In particular, a common goal was set in 2001 at European level to halve the road accident victims by 2010.
Different instruments have been set up in order to achieve this objective, such as education campaigns, use of driver
assistance systems in the vehicle and the design of safer transport infrastructure through an automated network in
constant communication with all elements of the road. The occurrence of an accident must take into account the key
variables that constitute the whole road system: the road, the driver, the vehicle and the environment. Among these,
a significant role is played by the driver and the vehicle. Based on the results of recent research on the causes of
accidents, human error comes into play in almost 93% of accidents and is the only cause of accidents in about three
quarters of cases (Commission of the European Communities Brussels, 2006).
The total number of accidents can be reduced through the safety systems installed in vehicles. However, it was
found that many "traditional" (or passive) safety measures (such as seat belts, airbags, etc.) are reducing their

*
Corresponding Author. Tel.: +39 06 5733 3416; fax: +39 065733 3441
E-mail address: [email protected].

1877–0428 © 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of the Organizing Committee
doi:10.1016/j.sbspro.2011.08.075
 Francesco Bella and Roberta Russo / Procedia Social and Behavioral Sciences 20 (2011) 676–686 677

effectiveness. It is therefore unlikely that they will bring further improvements in safety at reasonable cost. Passive
safety systems therefore have to be complemented by introducing on the market more advanced safety features in
vehicles (Commission of the European Communities Brussels, 2003), i.e. active and collaborative systems. With
regard to this, there are already on the market driver assistance systems in the vehicle, designed to provide the driver
with a valid support to assess non-evident risk situations, thus limiting the driving errors. In order to maximize their
effectiveness, these systems should be primarily shared by the driver, i.e. the system should provide warnings in
situations of “real” risk, for which the driver takes corrective action of his/her motion (e.g. deceleration). Too
frequent and too early alarms could be interpreted as “an annoyance” and ignored by the driver to the point of
bringing him/her to decide to turn off the device. In this case, the system would clearly not serve the purpose for
which it has been designed. In the design of active systems, it is therefore of utmost importance to find a proper
balance between not alerting too early, and yet to secure enough time of reaction, i.e. as to leave the driver the right
time and adequate space to avoid a possible collision or at least reduce the severity of the potential impact. The
present work was undertaken in this context, and it has the objective to analyze the driver’s behavior to contribute to
define driver assistance systems which are calibrated on the driver’s behavior and therefore can be readily accepted
by the driver.
Many advanced driver assistance systems (ADAS) exist and they allow investigating a wide number of
parameters and maneuvers (lateral control, longitudinal control, parking and reversing aids, vision enhancement,
driver monitoring, pre-crash systems, road surface or low-friction warning) (Lundgren and Tapani, 2006). As the
driving assistance systems act when it is still possible to avoid the accident, or significantly reduce its severity while
maintaining the high level of attention of the driver, the notice or the intervention in critical situations could help to
significantly improve road safety. This is very important because wrong driving behaviors, distracted driving and
excessive speed are among the main causes of accidents. We chose to focus on the analysis of driver behavior in
car-following as the majority of road accidents is between two or more vehicles. Furthermore, the most common
type is the front-side crash, followed by rear-end collisions. Therefore these are the situations of greatest risk.
In particular, in this study we analyzed tests conducted using the driving simulator of the Inter-University Research
Centre for Road Safety (CRISS) site in the University of Roma TRE. The driver behavior during car-following was then
considered in order to assess the accepted risk rear-end collisions. Various studies have shown that observations derived
from driving simulators are a reliable source of data to examine drivers’ behavior and the assessment of rear-end collision
risk (Alicandri, 1994; Bella, 2009; Farah et al.,2009; Jenkins and Rilett, 2004; Broughton et al. 2007), or to analyze
the efficiency of driver assistance systems (Cheng et al. 2002; Abe and Richardson, 2005; Jamson et al, 2008). That
because advanced-interactive driving simulators provide a high degree of realism, allow experiments to be conducted in
controlled conditions and assure the highest degree of safety for test drivers.
The study was developed through the following steps:
x analysis of methods and algorithms found in the literature on driver assistance systems;
x driving simulation study to record the driver behavior during car-following situations;
x analysis of data from simulated driving;
x development of a new algorithm based on driver behavior for a driver assistance system.

2. Background

Longitudinal collision avoidance technologies address rear-end collisions, which are usually due to the difficulty
for the driver to react to a sudden braking by the leading vehicle. A collision avoidance system operates in the
following manner: a sensor installed at the front of a vehicle constantly scans the road for vehicles. When one is
found, the system determines whether the vehicle is in imminent danger of crashing, and if so, a warning is issued,
or a collision avoidance maneuver is undertaken. The typical criteria for activation of collision avoidance are:
x Worst-case scenario: the system assumes that the lead vehicle could brake at full braking power at any time. In
essence, it maintains a “safe distance”, i.e. the minimum distance necessary to come to a stop in the event the
leading car suddenly brakes.
x Time-to-collision (TTC): the system determines the time required for two vehicles to collide if they continue at
their current speed and on the same path (assuming that VF is greater than VL).
The Figure 1 shows a car-following situation and the considered variables.
678 Francesco Bella and Roberta Russo / Procedia Social and Behavioral Sciences 20 (2011) 676–686

Figure 1 The car-following situation: variables of interest

As reported by the literature, several algorithms have been developed to activate the Collision Warning Systems
and in particular the two main types are based on kinematic and perceptual approach. Whatever the criterion of
activation, however, there are two major challenges: the system must be reliable and functional, secure and shared
by the driver (i.e. it should give few false alarms). In the following sections we analyze these two different
approaches, which are based on different assumptions and therefore operate in different ways.

2.1. Kinematic approach

The algorithms based on kinematic trigger alerts use the fundamental laws of motion. Combining the hypothesis
of the deceleration and reaction time with the current state of a vehicle, the algorithm determines a minimum
distance required to stop safely. When the vehicle is at a distance minor or equal to the safe distance from another
vehicle, an alarm is triggered. The kinematic approach starts the warning based on the criterion of the worst case.
From a thorough exam of the existing literature (Ararat et al., 2006; Brown et al., 2001; Jamson et al., 2008;
Kiefer et al., 1999; Kiefer et al., 2003; Kiefer et al., 2005; Lee and Peng, 2005; Martinez and Canudas-de-Wit, 2007;
Seiler et al., 1998; Zhang et al., 2006), we grouped different algorithms that evaluate on the basis of kinematic laws
of motion the critical warning distance (Rwarning) for Collision Warning (CW). These algorithms calculate a threshold
distance based on vehicle motion and the variables related to human characteristics (vehicle speed, acceleration,
delay in the human response, etc.). When the distance is smaller than the limit value calculated according to the type
of system used, an alarm is activated. The Table 1 shows the equations and terms most frequently used in CW
systems.

Table 1 Frequently used equations and terms in CW Systems

Meaning Form

Following vehicle stopping distance (RF) RF = 0.5 v2 F / (-αF), αF < 0, vF ≥ 0

Leading vehicle stopping distance (RL) RL = 0.5 v 2
L/ (-αL), αL < 0, vL ≥ 0
Reaction time margin (Rtd) Rtd = τvF, vF ,τF > 0
Range rate margin (RRR) RRR = τvrel , τ > 0 , vrel>0
Minimun range (Rmin) Rmin = costant > 0

where vrel = ΔV is the relative speed between the following and leading vehicles in m/s. For the other parameters
see Figure 1.
The main types of kinematic algorithms found in the literature are described below.
 Francesco Bella and Roberta Russo / Procedia Social and Behavioral Sciences 20 (2011) 676–686 679

2.1.1. The Mazda Algorithm

The algorithm developed by Mazda considers a hypothetical worst case. The scenario assumes that initially the
two vehicles maintain constant speeds: VL and VF. Subsequently, the lead vehicle starts to brake after time τ2 with a
deceleration rate αL, while the host vehicle starts to brake after an additional time τ 1 at deceleration rate αF, which
continues until both vehicles come to a full stop. The algorithm shown below calculates the minimum space required
to ensure that the scenario described above occurs without collisions.
R warning = f (vL,vF,vrel )= 0.5 [(v2F / αF) – (v2L /αL )] + vF τ1 + vrel τ2 + Rmin (1)
where Rmin is the minimum range (see Figure 1 and Table 1). The distance calculated by this formula must be
taken as a critical warning distance (Ararat et al., 2006). The values of the variables are reported in Table 2.

2.1.2. The Stop Distance Algorithm (SDA)

This algorithm defines a warning distance based on the difference between the stopping distances of the leading
and following vehicles. If the distance between the two vehicles is minor than the warning distance, an auditory
collision warning alarm is presented to the driver. The following describes the logic activating the alarm:
R warning = f (vL,vF ) = vF τdriver + (v2F/ 2αF) – (v2L / 2αL) (2)
where Wdriver is the driver’s reaction time. The values of the variables are reported in Table 2.

2.1.3. The PATH (Berkeley)Algorithm

This algorithm is a modified version of Mazda’s algorithm. The critical warning distance is given by:
R warning = f (vL,vF ) = 0.5 [(v2F / α) – (v2L /α)] + vF τ + Rmin (3)
According to this algorithm, the leading vehicle brakes at the maximum deceleration rate α, while the following
vehicle starts to brake after reaction time τ at the same deceleration. Rwarning is the minimum distance required to
avoid collisions until both vehicles come to a full stop. The values of the variables are presented in Table 2.

Table 2. Parameter values for algorithms

ΑF (m/s2) αL (m/s2) τ1 (s) τ2 (s) τdriver (s) τ (s) Rmin (m)

Mazda 6 8 0.1 0.6 - - 5
SDA 5 5 - - 1.5 - -
PATH 6 6 - - - 1.2 5

2.1.4. CAMP FCW project

Another very interesting model was proposed by Kiefer et al. (Kiefer et al., 1999; Kiefer et al., 2003; Kiefer et
al., 2005). Drivers were asked to execute last-second braking and steering maneuvers while approaching a surrogate
target lead vehicle. Based on the data recorded on field, the CAMP Forward Collision Warning (FCW) system was
defined. It determines the alert range necessary to assist the driver to avoid a potential crash, which is a function of
speeds and decelerations of the vehicles (leading and following) and total delay time. Total delay time was the
composite sum of three separate delay times: the interface delay time, the driver brake delay and the brake system
delay time. The interface delay time is defined as the time between when the crash alert criterion was violated and
when the crash alert was presented to the driver. This delay is assumed to be 0.18 s. The driver brake delay is
defined as the time between crash alert onset and when the driver triggered the brake switch. This delay was
assumed to be 1.50 seconds for surprise alerts (Olson and Sivak, 1996). The brake system delay time is defined as
the time between braking onset and vehicle slowing and is assumed to be 0.20 s. The total delay time is therefore
equal to 1.88 s.
R warning = f (vL,vF, αL, αF ,Total Delay Time) = ((vF – vL) (Total Delay Time) + 0.5 (αF – αL ) (Total Delay Time)2) +
((vF + (αF (Total Delay Time))) – (vL + (αL (Total Delay Time))))2 / (-2((- 1.61+ 0.668 (αL) - 0.807 (vF – vL) +
0.765)– αL))) (4)
680 Francesco Bella and Roberta Russo / Procedia Social and Behavioral Sciences 20 (2011) 676–686

2.2. Perceptual approach

With algorithms based on the perceptual approach, an alarm is triggered based on thresholds of perception. When
the human perceptual threshold is exceeded, a signal is activated to alert the driver. This approach is based on
thresholds of Time to Collision (TTC). The TTC is considered as one of the most widely used safety indicators and
a measure of a crash risk. It is defined as the time until a collision between two vehicles would have occurred if the
collision course and speed difference were maintained (Minderhoud and Bovy, 2001). The original definition was
coined by Hayward in 1972 (Hayward, 1972).
TTC = (xF – xL - lF) / (vF - vL) (5)
where lF is the length of the following vehicles; for the other variables see Figure 1.
The TTC is only calculated if the following vehicle’s speed is greater than the one of the leading vehicle (V F >
VL). In particular, low TTC indicate a higher risk of collision. Usually, thresholds value for the TTC are set to
trigger a warning, also called a critical value. The threshold value in the literature ranges between 5 and 2 seconds
(Minderhoud and Bovy, 2001). The main types of perceptual algorithms found in the literature are described below.

2.2.1. The Honda Algorithm

According to Honda’s algorithm, a warning is issued when the TTC is equal to 2.2 seconds. The distance is
calculated by means of:
R warning = f (vrel) = 2.2 vrel + 6.2 (6)
where 6.2 is an additional safety value.

2.2.2. The Honda’s Collision Mitigation Braking System (CMBS)

Another system by Honda, called Collision Mitigation Braking System (CMBS), is based on the evaluation of the
TTC. This system returns three different alerts. There is an initial phase of attention if the distance between the
following and leading vehicles exceeds the safety limit, estimated with a threshold of TTC equal to 3 seconds. When
this threshold is exceeded, an alarm starts to suggest the driver to take preventive measures to avoid collision. After
this, there are other phases of prevention and action that are activated if the distance continues to gradually decline,
i.e. TTC less than 2 seconds and less than 1 second respectively. In both cases the aim is to reduce the severity of
impact, but the collision will hardly be avoided.

2.2.3. The Hirst&Graham Algorithm

Many criticisms have been reported in the literature on the validity of the assumption of a constant threshold
value of the TTC to distinguish between safe and unsafe situations. It is often suggested to investigate if and how
this parameter varies depending on speed (Sultan and McDonald, 2003). In relation to the possible link between the
TTC and speed, it is worth mentioning a perceptual-based algorithm developed by Hirst and Graham (1997). They
suggest that a TTC-based warning should be increased with a speed penalty (SP) to achieve appropriately timed
warning. They suggest that this type of algorithm would minimize nuisance alarms. In essence, the algorithm
developed uses a threshold set at 3 seconds of the TTC, with an adjustment for the vehicle speed (SP) equal to
0.4905 m for each kilometer per hour of the following vehicle speed:
R warning = f (vrel,vF) = 3 dR/dt + 0.4905vF (7)
where dR/dt = ΔV is the relative speed between two vehicles.
 Francesco Bella and Roberta Russo / Procedia Social and Behavioral Sciences 20 (2011) 676–686 681

3. Method

A laboratory experiment using a driving simulator was developed in order to collect data on drivers’ car-
following behavior. The driving simulator of the Inter-University Research Centre for Road Safety (CRISS) in the
University of Roma TRE was used.

3.1. CRISS Driving Simulator

The CRISS simulation system is an interactive fixed-base driving simulator. It includes a complete vehicle
dynamics model based on the computer simulation Vehicle Dynamics Analysis Nonlinear. The model has been
adapted to run in real time and it has been validated extensively (Allen et al, 1998). Four computers are networked
and three interfaces (steering wheel, pedals and gearshift lever mounted on a real vehicle to reproduce a realistic
driving environment) constitute the hardware. A computer processes the equations of motion while the others
produce pictures. The driving scene is projected on three screens (see Figure 2 a). The usual field of view is 135°.
The resolution of the visual scene is 1024 × 768 pixels and the refresh rate is 30 to 60 Hz depending on the
complexity of the scene. The sound of the engine is reproduced through a sound system.
It is possible to implement different traffic scenarios: kinds of vehicles, numbers of vehicles for lane, speeds and
path of each vehicle. A set of parameters can be recorded to describe travel conditions (vehicle barycentre, relative
position in relation to the road axis, local speed and acceleration, steering wheel rotation angle, pitching angle,
rolling angle, etc.). The system also allows us to record the time and distance from the leading vehicle, as well as
time and distance from the vehicle in the opposite direction. All data can be recorded in intervals of time (a fraction
of a second) or space (a meter). Figure 2 shows the workstation for the design and implementation of the scenarios
in the simulator and a phase of a driving.

Figure 2. CRISS Driving Simulator: a) phase of the experiment b) workstation for the implementation of the scenarios

3.2. Test Alignment

An alignment of a two-lane rural road was designed and implemented in the driving simulator. The alignment
was 8.5 km long and had a cross-section 10.5 m wide (lane and shoulder widths were 3.75 m and 1.5 m
respectively). The alignment was divided into tangents with length ranging between 200 m and 1000 m and
horizontal curves made up of approach clothoid, circular curve and departure clothoid. The radii of the circular
curve ranged between 215 m and 1000 m. The traffic was introduced in both directions. Four scenarios were
simulated with a growing total traffic volume, ranging between 350 veh/h (scenario 1) and 900 veh/h (scenario 4), in
order to simulate four different volume-to-capacity ratios (N/C), ranging between 0.18 and 0.47. The speed of
vehicles changed in the four scenarios. In particular, in the opposite direction of the driver values between 85 and 70
km/h were given, while for vehicles traveling in the same direction of the driver the speed values were constant and
they varied between 75 km/h (in scenario 1) and 60 km/h (in scenario 4).
682 Francesco Bella and Roberta Russo / Procedia Social and Behavioral Sciences 20 (2011) 676–686

3.3. Procedure and participants

The study was carried out using dry pavement conditions in good state of maintenance, simulating the
characteristics of a medium-class car, both with regard to size and mechanical performance, with automatic gear-
changes. The data recording system was set to acquire all the parameters at spatial intervals of 5m.
The driving procedure was broken down into the following steps: 1) the driver is instructed on the duration of the
driving and the use of the steering wheel, pedals, and automatic gear; 2) the driver is trained in the driving simulator
on a specific alignment for approximately 10 min, so that he/she becomes familiar with the simulator’s control
instruments; 3) the execution of two test scenarios in the established sequence; 4) the driver gets out of the car for
about 5 minutes in order to re-establish psychophysical conditions similar to those at the beginning of the test and
fills in a form with personal data, years of driving experience, average annual distance driven; 5) the execution of
the two remaining test scenarios in the established sequence; 6) the driver fills in an evaluation questionnaire about
type (nausea, giddiness, daze, fatigue, other) and entity (null, light, medium, and high) of the discomfort perceived
during the driving. The sequence of the four scenarios was counterbalanced in order to avoid influences due to the
repetition of the same order in the experimental conditions. Drivers were instructed to drive as they would normally
do in the real world.
The simulation was performed by 32 drivers aged from 22 to 40, male (70%) and female (30%), with a driving
experience of at least 3 years and an average annual driven distance on rural roads of at least 2,500 km. From the
analysis of the questionnaire filled in by the drivers at the end of the test, it emerged that no participant experienced
any high or medium level of discomfort, 13 and 19 participants experienced light and null level of discomfort
respectively. Therefore no participant was excluded from the sample.

3.4. Data Processing

The data (longitudinal speed, distances between vehicles and their relative speeds) collected during the
simulations were analyzed to assess the conditions in which the driver adopts “evasive maneuvers” of car-following
such as overtaking maneuver or slow down to increase the spacing from the leading vehicle. The data were recorded
under these conditions:
x at the beginning of overtaking maneuver (only the accelerative maneuvers were considered);
x at the minimum spacing, during the car-following maneuver.
475 of these driving conditions were observed and then analyzed; 315 were overtaking and 160 were car-
following maneuver.
Figure 3 shows an example of the range between two vehicles in a car-following situation. In particular Ddriver1
and Ddriver3 refers to the first condition (overtaking maneuver; the vehicle which follows approaches the vehicle in
front, queues maintaining a constant speed and then overtakes), while Ddriver2 refers to the second one (minimum
spacing, during the car-following maneuver; the driver approaches vehicle in front, comes to a minimum distance
and then decelerates in order to increase the distance from it). These distances, called Ddriver, were used for the
analysis.

Figure 3. Example range between vehicles profile of a car-following situation

 Francesco Bella and Roberta Russo / Procedia Social and Behavioral Sciences 20 (2011) 676–686 683

In order to define an acceptable alert timing zone D driver should be added to a covered distance in the “drive brake
delay” which is defined as the time between crash alert onset and brake switch by the driver. In other words the
warning system should provide an alert Δt in advance compared to time when the driver is at the minimum spacing
from the lead vehicle. Based on experimental studies Kiefer ((Kiefer et al., 1999) suggest values of Δt equal to 1.18
s and 1.52 s (Figure 4).

Figure 4. Alert timing zone

The onset of the FCW crash alert must occur at any point within a crash alert timing zone, where this zone is
defined by “too early” and “too late” onset range cut-offs (or bounds) (Kiefer et al., 1999).

4. Results

4.1. Development of a new algorithm based on driver behavior

A statistical analysis was conducted to obtain a new collision warning algorithm that reflects the car-following
driver behavior during simulator testing. The multiple linear regression technique was used in model estimation. The
following model was found.
d= f (∆v,vF ) = 1,25∆v + 1,55 vF R2 = 0.89 (9)
where ΔV and vF are expressed in m/s.
The distance (d) given by the model can be considered as a threshold which should trigger an alarm system for
driver assistance. A such alarm is provided when the driver is in the middle of the crash alert timing zone (in other
words when 't is 1.35s). The independent variables used in the model were significant at the level of 5%. The
model is consistent as the distance to which the drivers take evasive maneuvers of the car-following condition,
increases as the speed of the driver (vF) and the relative speed between the following and leading vehicle (ΔV)
increases. It is a model with perceptual calibration system. It can be considered quite similar to the system of Hirst
and Graham (1997) (see equation 7) and but it provides different coefficients that reflect better the drivers’ behavior.
The new proposed algorithm was compared to algorithms reported in the literature. The performance of the
various models was studied by setting different values of speed of the driver (in a range between 100 km/h to 60
km/h). It should be noted that the range of the distances given by the different models is quite wide (Fig. 5). That
because the assumed hypotheses for developing the models are quite different .
The algorithm developed in this study resulted less precautionary (gives lower distances) than SDA for all values
of analysed speed (Fig. 5 a, b). As shown in Fig. 5, the new algorithm always provides higher values than the
684 Francesco Bella and Roberta Russo / Procedia Social and Behavioral Sciences 20 (2011) 676–686

CMBS-TTC 3s, Honda and CAMP.

For driver’s speed equal to 100 km/h (see Figure 5 a) the distances returned by the model are always higher than
those provided by the Hirst&Graham’s model; distances are higher than those returned by the PATH model for ΔV
up to 12 km/h, and by the Mazda model for ΔV up to 41 km/h. However, it must be noted that values of ΔV higher
than 30 km/h are rarely found on a rural two lanes road, in the conditions considered.
For driver’s speed equal to 60 km/h (see Figure 5 b), the new model is always more precautionary (gives higher
distances) than Mazda model. The same applies with regard to PATH model for ΔV up to 23 km/h (a more similar
pattern is observed afterwards), and to Hirst&Graham’s model for ΔV up to 36 km/h.

V=60 km/h
110

100

90

80

70
New
Distance

60 Mazda
50 PATH
SDA
40
Honda
30
CMBS-TTC 3s
20 Hirst&Graham

10 CAMP 1,88 s

0
10
0

15

20

25

30

35

40

45

50

55

60

65

ΔV

Figure 5. New algorithm compared with the warning distances of the literature: a) speed of driver (V) of 100 km/h; b) speed of driver (V) of 60 km/h
 Francesco Bella and Roberta Russo / Procedia Social and Behavioral Sciences 20 (2011) 676–686 685

5. Conclusions

The experimental study carried out at the CRISS driving simulator was aimed to analyze the driver behavior
during car-following maneuvers. Based on data collected in 475 car- following conditions a new collision warning
algorithm was defined. The distance given from the proposed model can be considered as a threshold which should
trigger an alarm system for driver assistance in order to advise the driver when he is in the crash alert timing zone.
The proposed algorithm reflects the real risk perception by drivers. Therefore it should minimize false alarms and
should help to avoid a potential collision.
Further studies are needed to validate the proposed algorithm (expand the database available in terms of drivers
and traffic types, several values of deceleration of the leading vehicle, etc.). It is planned in future research to
implement in the driving simulator a driver assistance system based on the algorithm proposed in order to test its
effectiveness.

References

Alicandri, E., 1994. The Highway Driving Simulator (HYSIM): the next best thing to being on the road. Public Roads 57 (3), 19-23.
Allen, R.W., Rosenthal, T.J., Aponso, B.L., Klyde, D.H., Anderson, F.G., Hougue, J.R. Chrstos, J.P., 1998. A low cost PC based driving
simulator for prototyping and Hardware-in-the-Loop Applications. SAE (Society of Automotive Engineers) paper 98-0222, Warrendale, PA.
Ararat Ö., Kural E., Güvenç B.A., 2006. Development of a collision warning system for Adaptive Cruise Control vehicles using a comparison
analysis of recent algorithms. Intelligent Vehicles Symposium 2006, June 13 – 15, Tokyo, Japan.
Archer J., 2005. Indicators for traffic safety assessment and prediction and their application in micro-simulation modelling: a study of urban and
suburban intersections. Doctoral Dissertation Royal Institute of Technology Stockholm, Sweden.
Bella F., 2009. Can the driving simulators contribute to solving the critical issues in geometric design? In Transportation Research Record:
Journal of the Transportation Research Board, No 2138, TRB, Washington, D.C., 120-126.
Broughton K. L.M, Switzer F., Scott D., 2007. Car following decisions under three visibility conditions and two speeds tested with a driving
simulator. Accident Analysis and Prevention 39, 106–116.
Brown T.L., Lee J.D., McGehee D.V., 2001. Human performance models and rear-end collision avoidance algorithms. Human Factors: The
Journal of the Human Factors and Ergonomics Society, Vol. 43, No. 3, 462-482.
Cheng B., Masahiro H., Takamasa S., 2002. Analysis of driver response to collision warning during car following. Society of Automotive
Engineers of Japan (JSAE) Review 23, 231–237.
Commission of the European Communities Brussels, 15.2.2006 COM (2006) 59 final. Communication on the Intelligent Car Initiative “Raising
Awareness of ICT for Smarter, Safer and Cleaner Vehicles”.
Commission of the European Communities Brussels, 15.9.2003 COM(2003) 542 final. Communication “Information and Communications
Technologies for Safe and Intelligent Vehicles” (SEC(2003) 963).
Farah H., Bekhor S., Polus A., 2009. Risk evaluation by modeling of passing behaviour on two-lane rural highways. In proceedings 88th Annual
Meeting Transportation Research Board. Washington D.C.
Genya A., Richardson A., 2005. The influence of alarm timing on braking response and driver trust in low speed driving. Safety Science 43, 639–
654.
Godbole D.N., Sengupta R., 1998. A method for design and specification of longitudinal controllers for vehicle automation. Proceedings of 8th
ITS America Meeting, Detroit, Michigan on Transportation technology for tomorrow.
Hayward J.C., 1972. Near miss determination through use of a scale of danger. Highway Research Record, 384, 24-34.
Hirst S., & Graham, R., 1997. The format and perception of collision warnings. In, Ergonomics and safety of intelligent driver interfaces, Y. I.
Noy (Ed.), 203–219.
Jamson A.H., Lai F.C.H., Carsten O.M.J., 2008. Potential benefit of an adaptive forward collision warning system. Transportation research. Part
C, Emerging Technologies, Vol.16c, No.4, 471-484.
Jenkins J.M., Rilett L.R. 2004. Application of distributed traffic simulation for passing behavior study. In Transportation Research Record:
Journal of the Transportation Research Board, 1899, TRB, Washington, 11-18.
Kiefer R.J., Cassar M.T., Flannagan C.A., LeBlanc D.J., Palmer M.D., Deering R.K., and Shulman M.A., 2003. Forward Collision Warning
Requirements Project: Refining the CAMP Crash Alert Timing Approach by Examining “Last-Second” Braking and Lane Change
Maneuvers Under Various Kinematic Conditions. (Report No. DOT-HS-809-574 ). Washington, DC: Department of Transportation
publication.
Kiefer R.J., LeBlanc D., Palmer M., Salinger J., Deering R., and Shulman M., 1999. Forward Collision Warning Systems: Development and
Validation of Functional Definitions and Evaluation Procedures for Collision Warning/Avoidance Systems. (Report No. DOT-HS-808-964).
Washington, DC: Department of Transportation publication.
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, Vol.37, Issue 2, 295-303.
Lee K., Peng H., October 2005. Evaluation of automotive forward collision warning and collision avoidance algorithms. Vehicle System
Dynamics Vol. 43, No. 10, 735–751.
686 Francesco Bella and Roberta Russo / Procedia Social and Behavioral Sciences 20 (2011) 676–686

Lundgren J., Tapani A., 2006. Evaluation of safety effects of driver assistance systems through traffic simulation. In Transportation Research
Record: Journal of the Transportation Research Board, 1953, TRB Washington, D.C., 81-88.
Martinez J.J., Canudas-de-Wit C., 2007. A safe longitudinal control for adaptive cruise control and stop-and-go scenarios. IEEE Transactions on
control systems technology.
Minderhoud M.M., Bovy P.H.L., 2001. Extended time-to-collision measures for road traffic safety assessment. Accident Analysis and Prevention,
Vol. 33, No. 1, 89-97.
Olson, P.L., and Sivak, M., 1996. Perception-response time to unexpected roadway hazards. Human Factors, 28, 91-96.
Seiler P., Song B., Hedrick J. K., February 1998. Development of a collision avoidance system. ITS Advanced Controls and Vehicle Navigation
Systems SAE Special Publications, pp 97–103.
Zhang Y., Antonsson E.K., 2006. Grote K. A new threat assessment measure for Collision Avoidance Systems. In Proceedings of the IEEE
Intelligent Transportation Systems Conference, 968–975, Toronto, Canada.

View publication stats