Safety Impacts of Queue Warning
in a Connected Vehicle Environment
Samaneh Khazraeian, Mohammed Hadi, and Yan Xiao
Queue warning systems (QWSs) have been implemented to increase
traffic safety by informing drivers about queued traffic ahead so that
they can react in a timely manner to the queue. Existing QWSs rely on
fixed traffic sensors to detect the back of a queue. It is expected that
if the transmitted messages from connected vehicles (CVs) are used
for this purpose, detection can be faster and more accurate. In addi-
tion, with CVs, delivery of the messages can be done with onboard
units instead of dynamic message signs and provide more flexibility on
how far upstream of the queue the messages are delivered. This study
investigates the accuracy and benefits of the QWS on the basis of CV
data. The study evaluated the safety benefits of the QWS under different
market penetrations of CVs in future years. Surrogate safety measures
were estimated with simulation modeling combined with the surrogate
safety assessment model tool. Results from this study indicate that a
relatively low market penetration—about 3% to 6%—for the congested
freeway examined in this study was sufficient for an accurate and reli-
able estimation of the queue length. Even at 3% market penetration, the
CV-based estimation of back-of-queue identification was significantly
more accurate than that based on detector measurements. The results
also found that CV data allowed faster detection of the bottleneck and
queue formation. Further, the QWS improved the safety conditions of
the network by reducing the number of rear-end conflicts. Safety effects
become significant when the compliance percentage with the queue
warning messages is more than 15%.
Rear-end collisions, generally caused by slow or stopped traffic,
are a main safety concern on freeways. About one-third of freeway
crashes are rear-end crashes (1). Recurrent bottlenecks, incidents,
and work zones are three main causes of slow queued traffic condi-
tions and consequently rear-end collisions. Queue warning systems
are designed to inform drivers about the queued traffic ahead so
that they can react in a timely manner. Daimler-Benz concluded
that 60% of rear-end collisions can be prevented if drivers had an
extra half second to react to slow traffic ahead (2). They also con-
cluded that 90% of rear-end collisions could be prevented if drivers
were given an additional second. Findings from the evaluation of
a number of queue warning systems (QWSs) showed reductions in
rear-end crashes that range from 14% to 44% (3–5). Such a reduc-
tion is most likely a function of many local and regional factors,
including geometry, local driving behavior, and other attributes of
the QWS deployment.
S. Khazraeian and Y. Xiao, EC 3730, Department of Civil and Environmental
Engineering, and M. Hadi, EC 3605, College of Engineering and Computing,
Florida International University, 10555 West Flagler Street, Miami, FL 33174.
Corresponding author: S. Khazraeian, skhaz001@fiu.edu.
Transportation Research Record: Journal of the Transportation Research Board,
No. 2621, 2017, pp. 31–37.
http://dx.doi.org/10.3141/2621-04
31
Peterson et al. investigated the accuracy and latency of the queue
warning and travel time estimation systems for a Minnesota I-94
smart work zone project (6). Portable dynamic message signs (DMSs)
were used to display the distance to the stopped or slow traffic on
the basis of traffic point detector data. Results showed that the accu-
racy of the distances to the queue were within 1 mi across differ-
ent queue lengths. The detectors were spaced 1 mi apart (7). The
maximum queuing detector error is expected to equal the detector
spacing distance.
Existing QWSs rely on fixed traffic sensors to detect the back of
queue (BOQ). Thus, the location of the BOQ cannot be detected
exactly, as indicated by the study of Peterson et al. mentioned above
(6). It is expected that if the transmitted messages from the con-
nected vehicles (CVs) are used for this purpose, the detection can be
faster and more accurate. In addition, with CVs, the delivery of the
messages can be done with onboard units instead of DMSs, provid-
ing more flexibility on how far upstream of the queue the messages
are delivered. In the future, vehicle responses to the queue warn-
ing can be automated when connected automated vehicles become
available. This study investigates the accuracy and benefits of the
QWS on the basis of CV data. The study evaluates the safety bene-
fits of the QWS under different CV market penetrations in the future
on the basis of safety surrogate measures estimated with simulation
modeling combined with the surrogate safety assessment model
(SSAM) tool.
Examples of Queue Warning Systems
Traffic point detection technologies have been used to detect queue
length. Video-based queue detection was implemented on the E313
freeway in Belgium. The detection algorithm read the speed and
occupancy measured by the video detection camera, and if the occu-
pancy was more than 50% and speed was less than 31 mph, a warn-
ing message was sent to the DMSs upstream of the incident (8).
The QWS implemented in Toronto, Ontario, Canada, on Catharine’s
Queen Elizabeth Way uses data from microwave sensor stations
and warns drivers about slow or stopped traffic. If the differential
speed between two microwave stations is high, drivers are advised
through DMSs (9). In a city in Denmark, a QWS is activated when
the speed obtained from traffic sensors is below 31 mph. After the
system activation, speed limits of 56, 44, and 31 mph are shown on
the successive DMSs upstream of the queue tail (4). The QWS in
Oslo, Norway, uses video detectors to detect the BOQ. If the speed
is below 19 mph and occupancy is higher than 30% and a speed
threshold is exceeded for more than 15 s, the warning is sent to
drivers through the DMS.
Pesti et al. proposed a point detector–based queue detection algo-
rithm and used Vissim microsimulation to evaluate the algorithm
32
(10). System parameters, such as speed thresholds, aggregation inter-
val, detector spacing, and portable DMS location, were examined.
A speed threshold of 35 mph, aggregation interval of 5 min, and
message update interval of 1 or 5 min were recommended by the
study (10).
Previous Research on CV-Based
Queue Warning
Balke et al. developed a speed harmonization algorithm and queue
warning algorithms, as part of the intelligent network flow opti-
mization prototype (11). Three types of queue warning algorithms
were proposed for the prototype: traffic management entity–based,
cloud-based, and vehicle-based. The traffic management entity
queue warning algorithm fuses the data obtained from the traffic sen-
sors and CVs to detect the BOQ and generate queue warning messages
through infrastructure signs and CVs. The process of determining
the BOQ is repeated every 5 s. In the cloud-based queue warning
algorithm, CVs send basic safety messages (BSMs), queued state
(Y or N), and mile marker location of the vehicle to the cloud using
cellular communication. The information is analyzed in the cloud,
and BSMs are assigned to sublinks according to the vehicles’
locations. Then, the BOQ information is sent with cellular com-
munications to the vehicles on the basis of their locations. In the
vehicle-based algorithm, each vehicle sends its mile marker loca-
tion and queued state (Y or N) every 1/10 s to surrounding vehicles
using dedicated short-range communications.
Dowling et al. (12) evaluated the effect of the prototypes of the
speed harmonization and queue warning algorithms developed by
Balke et al. (11), as described above. The speed harmonization and
queue warning prototypes were written in the Vissim COM inter-
face. Vissim microscopic simulation was used to model 8.5 mi of
the US-101 freeway in San Mateo, California. However, according
to the authors, the queue warning application could not be assessed
in the microscopic simulation because of the lack of information on
how drivers would react to the queue warning messages.
A small-scale demonstration was conducted by the Texas Trans-
portation Institute, Battelle, and the Washington State Department
of Transportation to equip 21 vehicles with CV technologies, trav-
eling on a 23-mi corridor of I-5 in downtown Seattle, Washington,
during the week of January 12, 2015 (13). The CV data were trans-
mitted and gathered with a bot cellular phone and dedicated short-
range communications. The study concluded that no loss of BSM
data was observed and there was no disruption in the algorithm
because of the loss of BSM data. The data capture, processing, and
delivery of messages to drivers took less than 10 s. The speed of that
process guarantees that drivers receive the queue warning message
1 mi in advance of the BOQ. The queue warning algorithm was
found to detect the BOQ 3 min sooner and could locate the BOQ
more accurately (0.5 to 1.5 mi farther upstream) than the road loop
detectors with 1-mi spacing.
Surrogate Safety Assessment Model
The implemented queue warning system is evaluated in this study
with the SSAM tool. The SSAM is a tool FHWA developed for
performing a safety analysis of vehicle trajectory data generated by
Transportation Research Record 2621
microsimulation software to calculate the frequency of different
types of conflicts, including crossing, lane-changing, and rear-end
conflicts, and severity indicators such as time to collision (TTC) and
postencroachment time (PET). The tool computes the safety of the
surrogate measures for each conflict and then generates a summary
of the surrogate measures, including the mean, maximum, mini-
mum, and variance. A conflict is a situation in which two vehicles
approach each other so that there is a risk of collision if they main-
tain their movements. TTC is estimated with the speed, position,
and future trajectory of the two vehicles. If the estimated TTC is
less than a threshold, the movement is identified as a conflict. The
conflict type is another output of the SSAM tool that determines
whether the conflict is a result of a rear-end, lane change, or crossing
movement (14).
FHWA studies indicate that of the SSAM outputs, frequency mea-
sures (number of conflicts) are more reliable than the severity indi-
cator measures such as TTC and PET (14, 15). Although a conflict
is counted when PET and TTC thresholds are met, the exact values
of the PET and TTC are not considered further beyond meeting the
thresholds when the frequency of the conflicts is determined. Sabra
et al. investigated the relationship between signal timing parame-
ters and the frequency of rear-end, angle, and lane change conflicts
resulting from SSAM output (16). The study reported the number of
conflicts only.
Methodology
Microscopic simulation modeling with the Vissim microscopic
simulation model is used for purposes of this research. This study
compares the accuracy of queue detection with point detectors to
detection with CV. Thus, this section first describes how the queue
is detected with those technologies in this study. A vehicle’s speed is
used as the measure to detect the queue formation and the BOQ. The
speeds are estimated according to point detector and CV data gener-
ated with simulation modeling. To simulate the CVs messages, the
trajectory conversion algorithm (TCA) tool, produced by FHWA
was used (17). The TCA can read vehicle trajectories from simula-
tion or real-world data and emulate the transmission of the BSM,
probe data messages, and European cooperative awareness mes-
sages. The study method assumes that the speed of the vehicles will
be estimated according to data collected from CVs with the BSM
specified in the SAE J2735 standards. BSM contains core data ele-
ments related to a vehicle, such as speed, acceleration, and position.
It is assumed that the BSM messages will be communicated to the
infrastructure with either dedicated short range communications or
low latency cellular communication technology.
MATLAB is used in this study to implement the bottleneck loca-
tion and BOQ detection algorithms–based CV and point detector
data generated by the simulation model, as described below. An inci-
dent is modeled in Vissim microsimulation to emulate a bottleneck.
Vehicle trajectories produced by Vissim under the incident condi-
tion are then input into the TCA tool to generate the BSM messages.
The freeway segment is decomposed to m 100-ft-long sub­segments.
The average speed of CVs in each subsegment is then calculated
for each analysis time interval, taken as 1 min in this study. First,
the bottleneck location is identified with the speed measurements,
generated by the TCA, by comparing each subsegment’s average
speed with the subsegment downstream of it. If the difference
Khazraeian, Hadi, and Xiao 33

is more than 30 mph, that segment is identified as a bottleneck
location (18).
To identify the BOQ, vehicles are sorted according to their posi-
tion compared with the bottleneck location, and the average speed of
vehicles in each segment is calculated. If the segment average speed
is below a threshold, the segment is considered queued. This study
uses the threshold recommended by Pesti et al., which is 35 mph as
the queue detection threshold (10). The threshold can be defined by
the analyst, possibly by including the use of zero speed (stop condi-
tions) as the threshold. The algorithm continues while the segment is
queued, and the first unqueued segment upstream of the bottleneck is
declared as the BOQ. This process is repeated each minute as shown
in Figure 1.
Finally, the performance of the CV-based BOQ detection is com-
pared with the Vissim-based detection and also with the queue esti-
mated according to point detection. The bottleneck location based on
point detectors is identified by comparing the upstream and down-
stream detector speeds. The detector-based BOQ algorithm is taken
from the Pesti et al. study, which estimates the location of the queue
with Equation 1 (10).
driver acceptance of the message advisory. Currently, information on
the driver’s responses to the disseminated queue warnings is limited.
This study performed a sensitivity analysis on the percentage of vehi-
cles reacting to the messages to determine the effects of this percent-
age. As more information becomes available from CV deployment,
it will be possible to input better estimates of the percentage into the
method to obtain better results. In the future, with the introduction of
CV automation, the response to queue warning messages will be set
automatically by the vehicle, and driver acceptance will become less
of a factor.
The vehicle’s trajectories produced by Vissim are fed to the TCA
tool to emulate BSM messages generated from the simulated vehi-
cles. The generated BSM messages are input into the bottleneck and
BOQ detection algorithms used in this study to investigate their per-
formance. The trajectories of the simulated vehicles were also input
into the SSAM tool to obtain safety surrogate measures to analyze
the benefits of the queue warning system.
Case Study

1 A well-calibrated network in Vissim was used to test the BOQ detec-

X = X DET (i ) + ∆X DET (1)
2 tion algorithms and the queue warning system of this study. The net-
where XDET (i) is the distance from the bottleneck location to the
speed detector i, which is the farthest detector from the bottleneck
location detecting the queue conditions from the bottleneck location,
and ΔXDET is the detector spacing.
The queue warning system is activated when the bottleneck is
detected and the queue starts growing. In this study, the queue warn-
ing effect is modeled by changing a certain percentage of the vehicle’s
speed upstream of the queue with the use of the Vissim COM inter-
face. It is assumed that the BOQ is detected by the CV data and the
queue warning message is shown dynamically at the DMS location
upstream of the BOQ (1 mi upstream of BOQ) or CV onboard units.
The proportions of vehicles changing speeds in response to the mes-
sages reflect the number of CVs equipped with onboard units and
work is a 20-mi segment of the I-95 southbound corridor in Broward
County, Florida. A one-lane blockage incident was introduced into
the traffic stream to generate a bottleneck location for the purposes
of this study. Since Vissim cannot model incidents directly, the inci-
dent was modeled with the creation of a signal head in one lane of
the freeway and the rubbernecking effect was modeled by changing
the car-following factors upstream of the incident location through the
Vissim COM interface.
The values of the car-following factors for the incident link
during the incident conditions were changed according to values
recommended in the literature. Knoop et al. explored driver behaviors
under incident conditions by looking at vehicle trajectories collected
by a helicopter from two actual incidents (19). Results showed that
driver reaction time, headway distribution, and capacity are affected

Vissim Trajectories
Under incident case (.fzp)

TCA
Running TCA and getting BSM

MATLAB

Sort BSM messages at time step t on the basis of their locations.

Calculate average speed for small segment i at time step t.

i=i+1
Is segment i average speed < 35 mph? Yes
No

This segment is back of queue at time step t. t=t+1

FIGURE 1   Back-of-queue estimation algorithm.

34 Transportation Research Record 2621

by the incident presence. A bimodal headway distribution, increase in
reaction time, and a 30% decrease in capacity were observed. Some
researchers found that incident effects can be modeled by changing
the Vissim car-following factors. A detailed description of the Vissim
car-following factors can be found in the Vissim user manual (20). A
study of the Vissim calibration parameters found that the CC0, CC1,
CC2, CC4, and CC5 parameters have the most effect on the capacity
of a freeway link (21). CC0 is the desired distance between stopped
cars. CC1 is the desired time headway that the driver wants to keep
from the leading vehicle. CC2 is the safety distance, which is the
minimum distance a driver keeps with the leading vehicle. CC4 and
CC5 are the parameters controlling the speed differences between a
following and a leading vehicle. Increasing CC0, CC1, and the abso-
lute value of CC4/CC5 results in capacity reduction. Gomes et al.
(22) modified three parameters, CC1, CC2, and CC4/CC5 to model
the capacity drops resulting from curvatures. The calibrated param-
eters for the studied freeway segment CC1, CC2, and CC4/CC5 were
1.7, 0.9 and −2/2, respectively (22). Rahman and Mattingly found
that the car-following values proposed in prior studies for incident
modeling produced better results relative to macroscopic measures
compared with the default values (23). Rompis et al. developed a
method for Vissim calibration based on kinematic queuing theory
(24). The incident was modeled by coding a traffic signal in Vissim.
On the basis of the reviewed study results, CC0, CC1, and CC2 val-
ues of 3.8, 1.5 s, and 7.5 m, respectively, were used to replicate the
actual freeway operation under incident conditions.
Vehicle measurements including vehicle speeds are assumed to
be continuously transmitted as part of BSM messages communi-
cated to the infrastructure. These messages can then be processed
by roadside units or at other infrastructure locations. As described
earlier, in this study, the TCA tool was used to generate BSM mes-
sages based on vehicle trajectories from the Vissim simulation
model. To test the methods presented in this study, a total of 40 sce-
narios with different CV penetrations and TCA seed numbers were
evaluated (four CV penetrations, and for each one, 10 TCA runs with
different seed numbers were performed). The purpose of having
multiple TCA runs is to randomly assign connectivity to vehicles
since the position of the CVs relative to the bottleneck location
affects the accuracy of the results. To confirm that the 10 TCA runs
are sufficient, the required number of runs at the 95% confidence
level was estimated with Equation 2.
σ2  t2
n= (2)
e2
where
σ = standard deviation of measurements,
t = associated t-value with 95% confidence interval (t-distribution),
and
e = acceptable measurement error (assumed to be 200 ft in this
study).
The calculated required number of runs for the 3% market pen-
etration was nine runs, and for 6%, 9%, and 15% it was two runs,
confirming that 10 runs was sufficient.
The generated BSMs were fed into the detection algorithms, and
the bottleneck location and BOQ were determined each minute. The
detected BOQ was compared with those detected by the Vissim
full trajectory, and the error percentage for each time step was
calculated with Equation 3.
( BOQ _Cik ) − ( BOQ _GTi )
error (i , k ) = 100  (3)
BOQ _GTi
where
error (i, k) = error percentage at time step i for kth TCA run,
BOQ_Cik = back of queue detected by CV data at time step i for
kth TCA run, and
BOQ_GTi = back of queue detected by full Vissim trajectory at
time step i.
The average error for different time steps for each run k was cal-
culated with Equation 4 and is shown in Figure 2. As shown in
this figure, the average error decreased with increasing market pen-

14

12
3% MP
10 6% MP
Error Percentage

9% MP
8
15% MP
50th percentile error
6

0
0 5 10 15
Market Penetration (%)

FIGURE 2   Average error percentages for each run (MP = market penetration of connected
vehicles in traffic stream).
Khazraeian, Hadi, and Xiao 35

TABLE 1   Error Percentage for Each Market the 100 runs for a given scenario) and median runs of the 10 TCA runs
Penetration Rate for the four CV penetrations versus those estimated with detector data
and simulation-based trajectories of all vehicles. As can be seen from
Market Penetration Error Range Median Error
Rate (%) (%) (%)
Figure 3, even the worst estimated queue lengths of the 10 TCA runs
are better than those estimated by the point detectors compared with
3 3.86–12.68 6.52 ground truth. With low market penetration (3%), however, although
6 2.25–5.37 3.72 the median of the runs produced accurate results at all time intervals,
9 2.03–6.30 2.94 the worst TCA runs produced errors at the beginning of the forma-
tion and dissipation of the queue, as shown for the 8th and 23rd min
15 1.36–3.22 2.40
in Figure 3a. These results reflect the lower number of vehicles in
shorter queues. However, with NHTSA’s expected mandate to install
CV technologies on all new vehicles, it is forecast that the CV market
etration. Median error for the 3%, 6%, 9%, and 15% CV penetra- penetration will be about 5% to 7% in the first year that this mandate
tions was 6.52%, 3.72%, 2.94%, and 2.36%, respectively. The error becomes effective, considering the rate at which new vehicles are
range for each market penetration is shown in Table 1. The detector- introduced in the traffic stream (25). The result from this study was
based error was calculated to be 49.53% compared with the Vissim also examined to determine how fast the bottleneck that starts the
simulation ground truth. queue buildup was detected. It was found that the bottleneck was
detected 4 min sooner with CV data compared with point detectors at
10

∑ error (i, k ) all market penetration rates, as can be seen from Figure 3a.
Once the queue is detected by the developed algorithm, the queue
Average_Error ( k ) = i =1
(4)
10 warning system is assumed to be activated by delivering messages
to the drivers ahead of the bottleneck. To implement the queue
Figure 3, a and b, shows the estimated queue lengths associated warning system, the driver’s response to the queue warning mes-
with the worst (worst run is the run that produced the largest error of sages needs to be known or estimated. Li et al. conducted a study

5,000
4,500
4,000
3,500
Queue Length (ft)

3% MP
3,000
6% MP
2,500
9% MP
2,000
1,500 15% MP
1,000 100% MP (ground truth)
500 Detector based
0
0 5 10 15 20 25 30
Time (min)
(a)

5,000
4,500
4,000
Queue Length (ft)

3,500 3% MP
3,000 6% MP
2,500
9% MP
2,000
1,500 15% MP
1,000 100% MP (ground truth)
500
Detector based
0
0 5 10 15 20 25 30
Time (min)
(b)

FIGURE 3   Estimated queue lengths by CV data versus detector-based and ground truth results
associated with (a) worst TCA run and (b) median TCA run.
36 Transportation Research Record 2621

TABLE 2   Statistical Analyses of Effects of QWS on Number of Rear-End Conflicts Using SSAM Model Results

Mean Difference Percentage Difference t-Value of Test to

Compliance with Mean Number Relative to Relative to No QWS Determine Significance
QWS (%) of Conflicts Significance No QWS (%) of Difference t-Critical

0 (no queue warning) 6,589.16 na na na na na

3 6,589.16 No 6.492 0.10 0.039 1.66
5 6,136.20 No 226.478 3.56 1.321 1.66
10 6,332.94 No 256.22 3.89 1.472 1.66
15 6,238.95 Yes 350.21 5.31 2.201 1.66
20 6,235.65 Yes 353.512 5.37 2.212 1.66
30 6,228.44 Yes 360.718 5.47 2.331 1.66
50 6,097.12 Yes 492.164 7.47 2.359 1.66
70 6,095.17 Yes 493.992 7.50 2.370 1.66

Note: na = not applicable.

of the effectiveness of portable changeable message sign messages
(“Road Work Ahead”) on reducing vehicle speed in rural highway
work zones (26). The work zone was located on US-36 and US-73
in Kansas. The study concluded that vehicle speeds were reduced
by 4.7 mph over a distance of 500 ft when the portable changeable
message sign was turned on. Richards et al. found, according to a
field study, that when a DMS was used as a speed control device in
work zones, vehicle speeds were reduced by 7 mph on average (27).
Zech et al. evaluated the effectiveness of three DMS messages on
vehicle speed and variance reduction in highway work zones (28).
The effective message reduced the speed by 3.3 to 6.7 mph. Dixon
and Wang found a 6- to 7-mph reduction in vehicle speed in the
immediate vicinity of DMS warning signs (29). Garber and Patel
concluded that vehicle speeds were reduced by 6 mph in response to
DMSs posted ahead of work zones (30). A similar study by McCoy
et al. reported that DMS activation resulted in a 4- to 5-mph reduc-
tion in vehicle speed and the percentage of drivers exceeding the
speed limit (45 mph) was reduced by 20% to 40% (31). This study
assumed that the messages were delivered with onboard units or
DMSs located about 1 mi ahead of the maximum queue length. The
complying drivers were assumed to reduce their speed by 10 mph,
and as a result, the average speed of the warning zone reduced by 3
to 10 mph depending on the compliance rate with the QWS.
The next step was to examine the safety effects of the QWS. The
trajectory file was imported into the SSAM tool to estimate the sur-
rogate safety performance measures. In this study, the trajectories
with and without queue waning were imported into the SSAM tool
to investigate the effect of the implemented queue warning system
on the number of rear-end conflicts. The tested compliance rates
were 3%, 5%, 10%, 15%, 20%, 30%, 50%, and 70%. The com-
pliance rate with the QWS was assumed to be the combination of
compliance with the DMS signs and the onboard units. One hundred
SSAM runs were performed for each compliance rate, and a t-test
with a significance level of .05 was conducted for each scenario.
Results of the t-test are shown in Table 2. The mean number of
rear-end conflicts for 100 SSAM runs for each compliance percent-
age scenario, as presented in the second column of the table, was
compared with the mean conflicts of the base conditions, which is
no queue warning scenario or zero compliance (first cell of first
column). The significance of the reduction in conflicts with different
compliance rates was determined (third column). The mean differ-
ence in the fourth column is the difference between the mean of
each scenario (with a specific compliance rate) and the mean of the
no queue warning scenario. As can be seen from Table 2, the reduc-
tion in the number of rear-end conflicts becomes significant when
compliance with the queue warning messages is equal to or more
than 15%. Most of the expected safety benefits are expected to be
achieved when a 50% compliance rate is reached, as the difference
in the percentage reduction in rear-end conflicts between 50% and
70% compliance rates is very small.
Conclusions
Results from this study indicate that a relatively low market penetra-
tion, about 3% to 6%, for the congested freeway examined in this
study, is sufficient for an accurate and reliable estimation of queue
length. It can be concluded that having 6% CVs in the traffic stream
allows the estimation of the BOQ location with less than 4% error on
average and a maximum error of about 5.4%. Even at 3% market pen-
etration, the CV-based estimation of BOQ identification is more accu-
rate than that based on detector measurements. It is also found that
CV data allow faster detection of the bottleneck and queue formation.
The CV-based algorithm can detect the start of a queue 4 min sooner
than the detector-based algorithm. Further, it is concluded that the
QWS improved network safety conditions by reducing the number of
rear-end conflicts. The safety effects become significant when com-
pliance with the queue warning messages is more than 15%. How-
ever, most of the expected safety benefits are expected to be achieved
when a 50% compliance percentage is reached; improvements with
increasing the compliance percentage above 50% are small.
References
 1. Vehicle- and Infrastructure-Based Technology for the Prevention of
Rear-End Collisions. National Transportation Safety Board, Washing-
ton, D.C., 2001.
 2. Safety Recommendation H-01-10 and H-01-11 N. National Transporta-
tion Safety Board. Washington, D.C., 2001.
  3. Study Tour Team. Innovative Traffic Control Technology and Practice
in Europe. FHWA, U.S. Department of Transportation, International
Technology Exchange Program, 1999.
  4. Wiles, P. B., S. A. Cooner, C. H. Walters, and E. J. Pultorak. Advance
Warning of Stopped Traffic on Freeways: Current Practices and Field
Studies of Queue Propagation Speeds. Texas Transportation Institute,
Texas A&M University System, College Station, 2003.
Khazraeian, Hadi, and Xiao
 5. Ullman, G. L., V. Iragavarapu, and R. E. Brydia. Safety Effects of
Portable End-of-Queue Warning System Deployments at Texas Work
Zones. Transportation Research Record: Journal of the Transpor-
tation Research Board, No. 2555, 2016, pp. 46–52. https://doi.org
/10.3141/2555-06.
  6. Peterson, S., E. Minge, and C. Iverson. Queue Warning and Travel Time
Estimation Near a Work Zone. Presented at 92nd Annual Meeting of the
Transportation Research Board, Washington, D.C., 2013.
  7. ARTBA Work Zone Safety Consortium. Innovative End-of Queue
Warning System Reduces Crashes Up to 45%. 2015. https://www
.workzonesafety.org/publication/innovative-end-of-queue-warning
-system-reduces-crashes-up-to-45/.
 8. E313 Antwerpen to Hasselt, Belgium Case Study, 1998. http://www
.traficon.com/inhoud_1_4_3_1.lasso?Case_ID=58. Accessed Aug. 22,
2002.
  9. Alexander, W., and M. Chamberlain. ITS in Ontario: The COMPASS
System. Paper for land use and transportation graduate course, University
of Toronto, Ontario, Canada, 2002.
10. Pesti, G., C.-L. Chu, H. Charara, G. L. Ullman, and K. Balke. Simu-
lation Based Evaluation of Dynamic Queue Warning System Per-
formance. Presented at 92nd Annual Meeting of the Transportation
Research Board, Washington, D.C., 2013.
11. Balke, K., H. Charara, and S. Sunkari. Report on Dynamic Speed
Harmonization and Queue Warning Algorithm Design. Texas A&M
Transportation Institute, Texas A&M University System, College Sta-
tion, 2014.
12. Dowling, R., A. Skabardonis, J. Barrios, A. Jia, and B. Nevers. Impacts
Assessment of Dynamic Speed Harmonization with Queue Warning.
FHWA, U.S. Department of Transportation, 2015.
13. Stephens, D., T. Timcho, J. Schroeder, J. Brown, P. Bacon, T. Smith,
K. Balke, H. Charara, and S. Sunkari. Intelligent Network Flow Optimi-
zation (INFLO) Prototype Seattle Small-Scale Demonstration Report.
Intelligent Transportation Systems (ITS) Joint Program Office, Research
and Innovative Technology Administration, U.S. Department of Trans-
portation, 2015.
14. Gettman, D., L. Pu, T. Sayed, and S. Shelby. Surrogate Safety Assess-
ment Model and Validation: Final Report. FHWA-HRT-08-051. FHWA,
U.S. Department of Transportation, 2008.
15. Gettman, D., and L. Head. Surrogate Safety Measures from Traffic
Simulation Models. FHWA-RD-03-050. FHWA, U.S. Department of
Transportation, 2003.
16. Sabra, Z. A., D. Gettman, R. D. Henry, and V. Nallamothu. Balancing
Safety and Capacity in an Adaptive Signal Control System—Phase 1.
FHWA-HRT-10-038. FHWA, U.S. Department of Transportation, 2010.
17. Trajectory Conversion Algorithm–VISSIM Software User Manual.
FHWA, U.S. Department of Transportation, 2014.
18. Khazraeian, S., M. Hadi, and Y. Xiao. Detection of Freeway Incidents
Based on Vehicle Acceleration Measurements Using Connected Vehicle
37
Data. Presented at 95th Annual Meeting of the Transportation Research
Board, Washington, D.C., 2016.
19. Knoop, V. L., H. J. Van Zuylen, and S. P. Hoogendoorn. Microscopic
Traffic Behavior Near Incidents. In Second Sino-Dutch Joint Workshop
in Transportation and Traffic Study, Shanghai, China, 2009.
20. VISSIM 5.40—User Manual. PTV AG, Karlsruhe, Germany, 2012.
21. Woody, T. Calibrating Freeway Simulation Models in VISSIM. Univer-
sity of Washington, Seattle, 2006.
22. Gomes, G., A. May, and R. Horowitz. Calibration of VISSIM for a Con-
gested Freeway. California PATH Program, Institute of Transportation
Studies, University of California, Berkeley, 2004.
23. Rahman, Z., and S. P. Mattingly. Macroscopic Evaluation of Incident-
Induction Driver Behavior Changes. In Symposium Celebrating 50 Years
of Traffic Flow Theory, Portland, Ore., 2014.
24. Rompis, S. Y., F. G. Habtemichael, and M. Cetin. A Methodology for
Calibrating Microscopic Simulation for Modeling Traffic Flow Under
Incidents. In IEEE 17th International Conference on Intelligent Trans-
portation Systems (ITSC), Qingdoa, China, 2014.
25. Wright, J., J. K. Garrett, C. J. Hill, G. D. Krueger, J. H. Evans, S.
Andrews, C. K. Wilson, R. Rajbhandari, and B. Burkhard. National
Connected Vehicle Field Infrastructure Footprint Analysis. Final Report
FHWA-JPO-14-125. FHWA, U.S. Department of Transportation, 2014.
26. Li, Y., Y. Bai, and U. Firman. Determining the Effectiveness of PCMS
on Reducing Vehicle Speed in Rural Highway Work Zones. Presented
at 89th Annual Meeting of the Transportation Research Board, Wash-
ington, D.C., 2010.
27. Richards, S. H., R. C. Wunderlich, and C. L. Dudek. Field Evaluation
of Work Zone Speed Control Techniques. Transportation Research
Record, No. 1035, 1986, pp. 66–78.
28. Zech, W. C., B. Mohan, and J. Dmochowski. Evaluation of Mes-
sages on Changeable Message Signs as a Speed Control Measure in
Highway Work Zones. Practice Periodical on Structural Design and
Construction, Vol. 13, No. 1, 2008, pp. 11–18. https://doi.org/10.1061
/(ASCE)1084-0680(2008)13:1(11).
29. Dixon, K., and C. Wang. Development of Speed Reduction Strategies for
Highway Work Zones. Final Report FHWA-GA-02-9810. FHWA, U.S.
Department of Transportation, 2002.
30. Garber, N. J., and S. T. Patel. Control of Vehicle Speeds in Tempo-
rary Traffic Control Zones (Work Zones) Using Changeable Message
Signs with Radar. Transportation Research Record, No. 1509, 1995,
pp. 73–81.
31. McCoy, P. T., J. A. Bonneson, and J. A. Kollbaum. Speed Reduction
Effects of Speed Monitoring Displays with Radar in Work Zones on
Interstate Highways. Transportation Research Record, No. 1509, 1995,
pp. 65–72.
The Standing Committee on Intelligent Transportation Systems peer-reviewed
this paper.