LANE DEPARTURE WARNING SYSTEM RESEARCH AND TEST DEVELOPMENT

Frank S. Barickman
National Highway Traffic Safety Administration
Larry Smith
Robert Jones
Transportation Research Center, Inc.
United States of America
Paper number 07-0495

ABSTRACT number of fatalities and injuries associated with

SVRD [3]. Based on 1991 General Estimate System
According to Traffic Safety Facts 2005 [1], single- and Fatal Accident Recording System data, Wang
vehicle crashes resulted in over 58% of all vehicular and Knipling reported that SVRD crashes accounted
fatalities on the nation’s roadways during that year. for almost 1.3 million of the 6.11 million police
Of these fatal crashes, almost 15,000 occurred either reported crashes and about 37.4% of all fatal vehicle
off of the roadway or on the shoulder. The National crashes [2].
Highway Traffic Safety Administration has
recognized that technologies such as electronic Since that time, NHTSA has continued to study the
stability control and other emerging safety SVRD problem to increase the understanding of the
technologies can potentially reduce a great number of crash problem and to help foster the development of
these fatal crashes. this crash avoidance technology. In the mid and late
1990s, NHTSA developed performance guidelines to
One emerging technology that the National Highway eliminate and mitigate road departure crashes [4].
Traffic Safety Administration believes may have This work ultimately specified performance
great potential to save lives is lane departure warning. guidelines for both a LDW system and a curve speed
These systems assist the driver by providing a warning (CSW) system. Pomerleau also estimated
warning (passive or active) that their vehicle is about that approximately 10% of all passenger vehicle road
to depart the road lane. The actual number of lives departure crashes can be prevented with LDW
saved would depend upon the effectiveness of the technology [4].
lane departure warning system.
In a more recent effort as part of the Intelligent
This paper will discuss both the past and present Vehicle Initiative program, NHTSA completed a road
research that has been conducted by the National departure crash warning system (RDCWS) field
Highway Traffic Safety Administration. It will give operational test (FOT). The RDCWS FOT studied
a general overview of the performance and potential both a lateral drift warning system and a CSW system
safety benefits of the technology. Information on the in an operational test environment. The study
type of sensors and performance testing to evaluate observed 78 subjects’ driving behavior for 1 month: 1
lane departure warning systems will be presented, week baseline without the RDCWS enabled and 3
including examples of them. Data from past field weeks with the RDCWS enabled. The study found
operational tests and test track research documenting that the LDW function had three major influences on
system performance will be shown. the subjects [5]:

INTRODUCTION • Turn signal usage per mile driven increased by

9%. (Note, that the system suppressed warnings
The National Highway Traffic Safety Administration when the turn signal was activated.)
(NHTSA) has long recognized that single-vehicle • The standard deviation of lane position was
road departure (SVRD) crashes lead to more fatalities decreased significantly.
than any other crash type [2]. Lane departure • Vehicles returned to the lane of travel quicker
warning (LDW) was a key technology identified at after being issued an imminent alert as compared
the start of the Intelligent Vehicle Highway System to lane excursions during the baseline week.
(IVHS) program that could potentially reduce the

Barickman 1
As part of the FOT, the Volpe Center served as the specifications, and operational requirements for
independent evaluator for the project. In a LDW technology. In some cases, these procedures
presentation about the preliminary RDCWS findings and/or guidelines have been developed for specific
[6], it was reported that the RDCWS with full programs such as the RDCWS FOT, but in general
deployment and availability could result in 34,000 to many of the concepts they test or specify are very
82,000 fewer lane departure crashes. similar. The following list of performance tests was
reviewed:
Crash statistics show that over time, SVRD crashes
have remained the largest category of crashes that 1. Recommendations for Objective Test Procedures
result in fatalities. From the crash problem for Road Departure Crash Warning Systems [7]
description described by Wang and Knipling in 1994, 2. ISO/CD17361 Lane Departure Warning Systems
a similar problem remains today as documented by [8]
Traffic Safety Facts 2005 (approximately 40%). 3. Development of Test Scenarios for Off-Roadway
Data from the FOT demonstrates that this technology Crash Countermeasures Based on Crash
has the potential to reduce SVRD crashes. Statistics [9]
4. Run-Off-Road Collision Avoidance Using IVHS
PERFORMANCE TEST EVALUATION Countermeasures [4]
5. Concept of Operations and Voluntary
LDW can be effective in preventing lane departure Operational Requirements for LDWS On-board
crashes because the technology can prevent the Commercial Motor Vehicles [10]
vehicle from departing the lane by either warning the
driver or actively controlling the vehicle. Similarly, Items 1 and 2 in the above list specify detailed test
ESC is effective in preventing lane departure crashes procedures on how LDW performance testing can be
because the technology can either limit a vehicle’s conducted. A variety of test scenarios, conditions,
tendency to oversteer, thus preventing it from and detailed procedures are defined. Item 3
spinning out of control or mitigate excessive recommends a series of more abstract tests that can
understeer, thereby preventing a vehicle from be performed to assess LDW performance based on
“plowing” off the road in a sharp curve. Whereas developing tests from statistical crash data. Najm
ESC systems assist drivers who do too much steering suggests that 96.3% of all road departure crashes
in a lane departure event, LDW systems assist drivers stem from just six conflict scenarios [9]. Items 4 and
that do not steer by alerting them. These systems 5 do not necessarily define performance tests, but
function at opposite ends of the crash spectrum. provide performance specifications and operational
requirements that should be met by an LDW system.
LDW systems have recently been introduced as
original equipment on late model vehicles in Japan, A detailed summary comparing and contrasting the
Europe, and North America. Unfortunately, it is still above listed efforts is beyond the scope of this paper,
too early to support any traditional benefit analysis but there are many common concepts that are
(crashes before technology vs. crashes after recommended to be tested. They all indicate that an
technology) due to low market penetration. LDW system should be able to function using
However, many have been trying to understand if different roadway delineations. These include both
benefits can be estimated through performance tests solid and dashed lines, yellow and white lines, and
and objective test development. raised pavement markings. They all recommend (or
suggest demonstrating via a test) that LDW warnings
In an effort to understand how LDW systems can should be issued for straight roads (>1000m radius of
potentially reduce SVRD crashes, NHTSA has been curvature) and curves (various radius of curvature
studying current LDW technology. For an LDW 50m to 1000m) within some time frame (or distance)
system to reduce crashes, it must operate at a certain of the lane marking at a variety of road departure
level of performance under varying conditions. The rates. The lateral departure rates vary from 0.1 to 0.8
purpose of this testing was to identify what objective m/s. Some of the other common concepts include a
test procedures could be used to measure the minimum operational speed (and/or test-specific
performance of LDW technology. speeds), tests to determine if the warning is
suppressed by turn signal usage, and environment
Exisiting Objective Performance Tests conditions for the tests.

During recent years, NHTSA researchers and others

have been developing performance tests,

Barickman 2
Test Vehicle and Measures TABLE 1.
Primary measures collected by the onboard data
For this testing, a passenger car was instrumented for acquisitions for testing.
data collection. The test vehicle was purchased with
original equipment (OE) lane departure warning System Measure Units Sample
system (LDW) that provided an audible and visual Description Rate
warning when the vehicle departs the lane. Also OE Departure Flag On/Off 30 Hz
included on the platform were an aftermarket (AM) LDW
LDW and a low-cost lane position measuring system AM Lateral Position Meters 5 Hz
(LPMS) [11]. LDW
AM Lane Width Meters 5 Hz
Both the OE and AM LDW systems use a forward LDW
looking video camera. Both systems issue auditory AM Lateral Velocity M/sec 5 Hz
and visual warnings to the driver to indicate lane LDW
departure. For this study, a detailed analysis of the AM Line Type Solid / 5 Hz
user interface was not appropriate. The output signals LDW Dashed /
were used as a means to indicate lane departure Unknown
electronically. The primary measures that were / None
collected are defined in Table 1. AM Lateral Position Percent 5 Hz
LDW Confidence
Raw measurement data were not available from the AM Warning Flag On/Off 5 Hz
OE LDW sensor. Derived measures such as warning LDW
time onset and lane line crossing had to be LPMS Lateral Dist to Left Meters 30 Hz
determined by fusing the OE LDW departure flag Left Line
(i.e. data channel marker) with other data. To LPMS Lateral Dist to Meters 30 Hz
compute warning time measures, time synchronized Right Right Line
video data were manually compared to the onset of
GPS High Accuracy Northings 10 Hz
the departure flag from the OE LDW. Other metrics
Position Position and
for the OE LDW were calculated by comparing the
Eastings
data from the other two sensors and/or the event
POI Point of Interest On/Off 30 Hz
button and monitoring the output response of the OE
Button (Experimenter
LDW system. Unfortunately, the ability to determine
Flag)
if the LDW is tracking the roadway line (availability)
cannot be completely assessed this way, but positive Video Left Down N/A 30 Hz
Left Looking Video
warning rates can be calculated (i.e. if we know a
lane line boundary was crossed, did the OE LDW Video Right Down N/A 30 Hz
warn or not?). Right Looking Video
Video Forward Looking N/A 30 Hz
Derived performance measures for the AM LDW Fwd Video
were calculated using the lateral position and lane
width channels as measured from the sensor. Lane
departures and warning times were calculated by Test Track Testing
comparing the lane bust measure to the AM LDW
warning flag. Data from the point of interest (POI) Performance testing for each system was conducted
button and other sensors were also compared to at the Transportation Research Center, Inc. (TRC) in
ensure that a lane bust actually occurred. For East Liberty, OH. Tests were conducted to assess
consistency, warning times were also compared how the systems generated warnings on both straight
manually to the video data. Availability was road segments and curves.
measured by monitoring the lane position confidence
channel. The first test was conducted on the straight section of
the Winding Road Course (WRC) at the TRC. This
test is very similar to the ISO repeatability test and
the NIST lateral drift on a straight road test. The
purpose of the test is to assess when warnings are
given with respect to departing the lane and how
repeatably the warnings are issued. To conduct the

Barickman 3
test, cones mark two different approach angles approaching the curve, the vehicle is accelerated to
leading up to a lane line. Using GPS measurements, 74 KPH. While in the curve, lane changes are
results are recorded by comparing the vehicle performed at an approximate lateral velocity of 0.3
position at the time of the warning to the position of m/sec and 0.8 m/sec. The exact distance from the
the painted road marking. painted line to the outside edge of the vehicle at the
time the LDWS alarm sounded was determined from
A rectangular course was marked 188m long by 3.6m GPS data.
wide, with one long edge of the rectangle being a
solid painted line as can be seen in Figure 1. Cones
were placed on the solid line at the entry, 54m, and
188m from the entry point. An additional cone was
placed 3.6m from the painted line to denote the width
of the course. The driver was responsible for
aligning the cone 3.6 m out from the painted line with
one of the cones at 54m and 188m, depending on
desired approach rates. Two calculated angles were
used to achieve the two approach rates of 0.3 m/s and
0.8 m/s at the controlled vehicle forward speed of 74
KPH. The exact distance from the painted line to the
vehicle at the time the LDWS alarm sounded was
determined from GPS data.

Figure 2. Layout of the curve lateral drift warning

3.6m scenario.

In the ISO test document, a false alarm test is

54m described. The false alarm test was conducted on the
188m skid pad at the TRC. Straight lanes with painted lines
approximately 2km long and 3.6m wide are available.
The test was conducted with the car driven directly
Figure 1. Layout of the straight lateral drift
down the center of the lane. No lane crossings are
warning scenario (not to scale).
performed. The objective of the test is to ensure that
no false alarms are generated.
The purpose of the second test was to determine the
Performance Testing Results
timeliness and repeatability of the warning during a
slow drift while in a curve. A figure displaying the
The summary results from this testing can be seen in
general test scenario is shown in Figure 2. This is
Table 2. Overall, both the OE LDW and the AM
similar to the ISO warning generation test and the
LDW systems were able to perform quite well in
NIST curved road lateral drift test. The ISO
these tests.
document prescribes that this test be performed in a
curve of radius 500m ±50m. No such curve was
One problem discovered during testing was that the
found in any available test facilities. The warning
AM LDW was not operating in a warning-enabled
generation test was attempted on a curve with a
state during the lateral drift test in a curve. Although
radius of 110m, the largest un-banked curve available
the sensor was functioning (i.e. lateral position was
on TRC property for this test.
being output) through the curve, the warnings were
suppressed because the initial approach did not have
The objective of this test was to achieve two different
lane lines. If the AM LDW senses that there are no
approach rates relative to the lane markings, in two
lane lines present for some period of time, the system
different directions through the curve, and to depart
enters a mode where warnings are suppressed. Once
the roadway on both the left and right side of the of
the system senses good quality lines for some time
lane. On a straight section of the roadway

Barickman 4
period, it automatically enables itself and is able to Repeatability was a little more variable than the
present warnings to the driver. In the real world, this straight lateral drift tests. It is believed that the
is done to prevent false alarms; however, from a test variability was caused by the test driver since it is
standpoint, this can be a problem when using a difficult to create the lane departure scenario in the
limited area. same manner on a curve (i.e., its harder to judge
where you cross the lane boundaries on a curved
TABLE 2. section of road vs. crossing a lane line while driving
Results of the performance testing conducted on straight.). Warnings were issued but were sometimes
the test track. outside of the ISO set +/- 30cm zone for each test
group.
OE LDW
# Description Low Lateral High Lateral Both systems were able to pass the ISO false alarm
Velocity Velocity test. This test is very easy to implement and run, but
it may be too simple to yield valuable data. Neither
L R L R
of the systems tested issued a false alarm (i.e., a
Straight warning from the LDWS without a lane departure or
1 Lateral Drift Pass Pass Pass Pass near-departure).
Warning
Curve Lateral Functional Testing
2 Drift Pass Pass Pass Pass
Warning Functional testing was performed to determine how
False Alarm the systems functioned under real-world road
3 Pass
Test conditions. This testing is similar to what Najm
AM LDW describes as system robustness testing. The tests are
# Description Low Lateral High Lateral performed on roads that are very similar to the types
Velocity Velocity of roadways described in the crash statistics. Since
L R L R the tests are conducted on public roadways, the
external test conditions cannot be tightly controlled,
Straight but they do provide a reasonable amount of
1 Lateral Drift Pass Pass Pass Pass variability that may be experienced in the real world.
Warning
Curve Lateral Functional testing was conducted on State roadways
2 Drift N/A N/A N/A N/A around the Marysville, Ohio area. The roads have a
Warning posted speed limit of 72-88kph, are non-freeway /
False Alarm two lanes, rural, and mostly straight with some
3 Pass
Test curves. The road markings appear to be in good
condition based on human visual perception. On the
right hand side of the road, the edge is delineated by
Both systems were able to correctly generate a constant white line. The left or center line of the
warnings during the straight lateral drift warning test. roadway is delineated by yellow solid and/or dashed
Warnings were issued within the given window lines. The road can further be characterized by
specified by the ISO test procedure under both lateral mentioning that the surroundings are mostly
drift rates. The alerts were issued within the ‘on agricultural and sparsely populated with rural
time’ rating as calculated by the NIST test procedure. housing.
They were issued prior to the latest warning line and
after crossing the earliest warning line determined by The test consisted of multiple drives over time. The
the lateral drift velocity. Finally, warnings were testing took place over multiple days and is done at
issued in a repeatable manner by both systems during different times of the day. During each drive, the
all tests. experimenter would regularly but randomly depart
the roadway as many times as they could on both the
The OE LDW system was able to pass the curve left and right sides of the road. The experimenter
lateral drift warning tests. Warning generation tests would indicate a road departure by pressing the POI
were within the window of the pass criteria set by the button every time the vehicle departed the lane. Data
ISO test procedure. Warnings were issued prior to were recorded both manually and electronically,
the latest warning line and after crossing the earliest recording if the LDW system(s) issued a warning to
warning line determined by the lateral drift velocity. the driver.

Barickman 5
One of the important aspects of functional testing is TABLE 3.
to negate environmental conditions over time. To Results of the functional testing conducted on
negate environmental conditions, tests using the same public roadways.
roadways were conducted over multiple days, times,
weather, and lighting conditions. Tests were also OE LDW
conducted using a “double-back” route, where the Segment Description
Depart Warn %
route return trip is the same route but in the opposite
direction, thus having the sensor face 180 degrees TRC
A 106 105 99.1%
from its initial trip. It is believed that the Property
environmental effects are negated using this method TRC Gate to
B 501 476 95.0%
because performance can be shown over a period of Raymond
time verses any one instantaneous moment. The fact Raymond to
C 287 284 99.0%
that weather, traffic, sunlight, etc. are constantly SR 31
changing can be negated if performance is SR 31 to SR
D 520 277 53.3%
consistently poor or good over a given section of 4
roadway. SR 4 South
E 449 431 96.0%
of SR 347
Functional Testing Results Totals 1863 1573 84.4%

The results of the functional testing are displayed in AM LDW

Segment Description
Table 3. The results are for a total of 12 test drives. Depart Warn %
At first look when evaluating the overall performance TRC
A 39 32 82.1%
of both systems, the results are comparable with both Property
systems performing in the 80 – 85% range. One TRC Gate to
important note is that the systems both operationally B 441 369 83.7%
Raymond
perform differently. This is evident in the number of Raymond to
departure attempts. The OE LDW is capable of C 264 167 63.3%
SR 31
warning the driver constantly when speeds are over SR 31 to SR
70kph. The AM LDW is not capable of warning D 457 419 91.7%
4
constantly. The AM LDW system suppresses SR 4 South
warnings for 5 seconds after it issues a warning. This E 381 286 75.1%
of SR 347
limits the overall number of departures that can be
accomplished during the same segment. This Totals 1582 1273 80.4%
operational difference also makes the AM LDW
departure attempts a subset of the OE LDW departure
attempts. DISCUSSION

Looking at the individual segments, performance Unfortunately, detailed data for the OE LDW were
differences become more obvious. The OE LDW not available for this testing. Only the basic inputs
system performs above 95% of the time on every (we departed a lane) and outputs (the LDW system
segment but one, which brings down its overall warned) were known for testing. If other data such as
average. The AM LDW does not perform as high as lateral position within the lane, lane width, line
the OE LDW but never performs lower than 63% marking type, and measurement confidence were
(10% higher than the worst OW LDW performance). known, a better understanding of why the OE LDW
performed poorly during section “D” of the
The other interesting observation from the data is that functional test might be known. Looking at the
the OE LDW’s worst performing section is the AM performance from the AM LDW was not helpful
LDW’s best performing section. It is unclear as to since it seemed to perform the best in this section.
why this phenomenon was observed. Again, all of
these roadway segments had lane markings that Overall, looking at the performance of the AM LDW,
looked average or better and they all looked visually the data generally suggest that the sensor sometimes
very similar. Tests were also conducted using both had trouble tracking the roadway markings. This was
systems at the same time. Since the AM LDW was indicated in the data as either low confidence or the
able to perform quite well, it is hard to suggest that absence of a lane boundary being sensed. This has
there is a particular problem with this segment. been discussed by others as “availability”.

Barickman 6
 REFERENCES
Similar conclusions were found in the RDCW FOT
where they identified that availability was, perhaps, 1. National Center for Statistics and Analysis. Traffic
the most important issue in LDW. They found that Safety Facts 2005. U.S. Department of
lane marking quality, camera obstructions, roadway Transportations, National Highway Traffic Safety
contamination (water, glare, snow, salt, etc.), and Administration.
ambient lighting conditions can impede the ability of
the system to correctly track the lane. 2. Wang, J., Knipling, R. R. Single Vehicle Roadway
Departure Crashes: Problem Size Assessment and
CONCLUSIONS Statistical Description. DOT HS 808 113, March
1994.
Assessment of LDW systems is a challenge. There
are many external influences that can cause problems 3. Mironer, M., Hendricks, D. Examination of Single
and degrade the performance of the system. Vehicle Roadway Departure Crashes and Potential
Although it may be important to characterize the IVHS Countermeasures. DOT HS 808 144, August
functional characteristics of an LDW system, simply 1994.
completing performance tests on a test track may not
be enough to gain insight into the real-world 4. Pomerleau, D., Jochem, T., Thorpe, C., Batavia, P.,
effectiveness of an LDW system. From the results of Pape, D., Hadden, J., McMillian, N., Brown, N.,
this study, it is believed that existing objective test Everson, J., Run-Off-Road Collision Avoidance
procedures do not adequately characterize real-world Using IVHS Countermeasures. DOT HS 809 170,
performance. Both the OE LDW and AM LDW December 1999.
systems performed quite well during the test track
scenarios; however, both systems had various 5. LeBlanc, D., Sayer, J., Winkler, C., Ervin, R.,
problems when tested on public roadways. Bogard, S., Devonshire, J. Mefford, M., Hagan, M.,
Bareket, Z., Goodsell, R., and Gordon, T. Road
A functional performance test may provide better Departure Crash Warning System Field Operational
operational insight about the performance of an LDW Test: Methodology and Results. UMTRI 2006-9-1,
system. Using this methodology, external influences June 2006.
can be minimized and real world performance can be
measured. Since both systems essentially passed test 6. Wilson, B. Road Departure Crash Warning
track testing, it appears both systems are equal in System: Preliminary Results of the Independent
performance. However, when comparing data from Evalution. http://www.itsa.org/itsa/files/pdf/RDCW
the functional test, it becomes obvious that the two Volpe Evaluation.pdf. ITS America: Public Meeting
systems perform quite differently. on the Benefits of Advanced Crash Avoidance
Systems. April 2006.
The idea of a functional performance test is quite
new, and there are many problems with the concept. 7. Szabo, S., Norcross, R., Recommendations for
One challenge is to make this test repeatable so that Objective Test Procedures for Road Departure Crash
similar results can be obtained from any group of Warning Systems. NISTIR 7288, February 2000.
similar roadways. Another problem is that roadways
are constantly changing over time. Even using the 8. ISO Draft International Standard. Intelligent
same roadways, the results may differ with the same Transport Systems – Lane Departure Warning
system. A third challenge for this testing is Systems – Performance Requirements and Test
developing pass/fail criteria for the test. Is it Procedures. Working Document N123.33.
acceptable for an LDW system to perform above ISO/CD17361. Committee ISO/TC204/WG14. June
90% and then have a section where it performs at 2003.
only 50%? Or is it better to have a system that
performs at above 80% under all conditions? To help 9. Najm, W., Koopermann, J. Boyle, L, and Smith, D.
understand these issues and answer these questions, Development of Test Scenarios for Off-Roadway
additional testing needs to be completed. Crash Countermeasures Based on Crash Statistics.
DOT HS 809 505, September 2002.

10. Houser, A., Pierowicz, J., Fuglewicz, D. Concept

of Operations and Voluntary Operational
Requirements for Lane Departure Warning System

Barickman 7
(LDWS) On-board Commercial Motor Vehicles.
FMCSA-MCRR-05-005. July 2005.

11. Barickman, F., Stoltzfus, D. A Simple CCD Based

Lane Tracking System. Society of Automotive
Engineers. 1999-01-1302. March 1999.

Barickman 8