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

net/publication/306010686

Lane Departure Warning and Prevention Systems in the US Vehicle Fleet:

Influence of Roadway Characteristics on Potential Safety Benefits

Article  in  Transportation Research Record Journal of the Transportation Research Board · January 2016
DOI: 10.3141/2559-03

CITATIONS READS

30 571

3 authors:

John M Scanlon Kristofer D. Kusano

Waymo Waymo
24 PUBLICATIONS   530 CITATIONS    47 PUBLICATIONS   1,024 CITATIONS   

SEE PROFILE SEE PROFILE

Hampton Gabler
Virginia Tech (Virginia Polytechnic Institute and State University)
201 PUBLICATIONS   2,924 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Dynamic Rollover Occupant Protection (DROP) View project

All content following this page was uploaded by John M Scanlon on 02 November 2016.

The user has requested enhancement of the downloaded file.

 1 The Influence of Roadway Characteristics on Potential Safety Benefits of Lane Departure
2 Warning and Prevention Systems in the U.S. Vehicle Fleet
3
4 John M. Scanlon
5 Virginia Tech - Department of Biomedical Engineering and Mechanics
6 440 Kelly Hall, 325 Stanger Street, Blacksburg, VA 24061
7 Email:john91@vt.edu
8
9 Kristofer D. Kusano
10 Virginia Tech - Department of Biomedical Engineering and Mechanics
11 440 Kelly Hall, 325 Stanger Street, Blacksburg, VA 24061
12 Email:kusano@vt.edu
13
14 Hampton C. Gabler
15 Virginia Tech - Department of Biomedical Engineering and Mechanics
16 445 Kelly Hall, 325 Stanger Street, Blacksburg, VA 24061
17 Email:gabler@vt.edu
18
19
20
21 Word count: 4,488 words text + 7 Tables/Figures × 250 words (each) = 6,238 words
22
23
24
25
26 Submission Date: 11/14/2015
 Scanlon, Kusano, Gabler

1 ABSTRACT
2
3 Road departure crashes account for one-tenth of all crashes, but nearly one-third of all fatal crashes. Lane
4 departure warning (LDW) and lane departure prevention (LDP) active safety systems have the potential to
5 mitigate these crashes by warning the driver of a lane departure and automatically navigating the vehicle
6 back into the lane, respectively. However, the effectiveness of these systems is expected to be highly
7 dependent on roadway characteristics, such as lane markings and shoulder width. The objective of this
8 study was to quantify the influence of these roadway characteristics on the effectiveness of LDW and
9 LDP within the U.S. vehicle fleet. This study used a simulation case set of 478 real world drift out of lane
10 road departure crashes taken from NASS/CDS to simulate these crashes as if the vehicles had been
11 equipped with LDW or LDP. The simulations were then repeated as if (1) all of the roadways had lane
12 markings present, (2) the roadway shoulders were expanded (3.6 m), and (3) lane markings were present
13 and the shoulder widths were expanded. Given current roadway infrastructure, LDW and LDP were found
14 to potentially prevent 28%-32% of U.S. road departure crashes and 21%-28% of seriously injured drivers.
15 When lane markings were added to the roadways, LDW and LDP could prevent 32%-36% of crashes and
16 27%-31% of seriously injured drivers. When only shoulder widths were expanded, LDW and LDP could
17 prevent 50%-54% of crashes and 44%-48% of seriously injured drivers. When both lane markings were
18 present and the shoulders were expanded, LDW and LDP could prevent 72%-78% of crashes and 60%-
19 65% of seriously injured drivers. The findings of this study highlight the important influence of roadway
20 infrastructure on the performance of LDW and LDP, and should be considered by designers of these
21 systems and roadway policy makers.
22
23
24 Keywords: Road Departure Crashes, Roadway Infrastructure, Lane Departure Warning, Benefits Estimates
 Scanlon, Kusano, Gabler

1 BACKGROUND
2 Although comprising only one-tenth of all crashes, road departure crashes account for nearly one-third of
3 all fatal crashes in the United States (1). Lane departure warning (LDW) and lane departure prevention
4 (LDP) systems are vehicle-based active safety systems that have the potential to reduce the number of
5 crashes and seriously injured drivers that result from these road departures. LDW can alert the driver of a
6 lane departure, thereby allowing the driver time to steer the vehicle back into the lane markings. LDP works
7 by directly controlling the vehicle’s trajectory in the event of a lane departure.
8 Several previous studies have evaluated the potential crash and safety benefits of LDW and LDP through
9 the retrospective simulation of real world lane departure crashes as if the vehicles had been equipped with
10 one of these systems (2-10). Each of these studies predict a potential reduction in the number of lane
11 departure crashes if these systems were equipped throughout the entire vehicle fleet. Modifications to many
12 elements of LDW/LDP have been investigated in the past, including the type of warning that should be
13 delivered to the driver (11; 12), when the warning should be delivered (13; 14), how to reduce driver
14 annoyance to LDW alarms (15), and how the driver will adapt to these systems (16). However, no studies
15 have analyzed how roadway infrastructure may limit or enhance the performance of these systems.
16 The performance of LDW and LDP are expected to be dependent on roadway infrastructure. Specifically,
17 roads without lane markings and roads with narrow shoulders may restrict the ability of these systems to
18 successfully prevent road departures. Current LDW/LDP systems rely on vision-based sensing of lane
19 markings to detect a lane departure (17-21). Future technologies may be capable of detecting curbs or lane
20 edge, but current systems do not activate on roadways with missing lane markings (22; 23). Additionally,
21 shoulders provide additional time and space for the vehicle to be driven back into the lane. Absent or narrow
22 shoulders will restrict the performance of these systems. Although advancements in road edge detection
23 technologies may solve the issue of absent lane markings, universally widening road shoulders would prove
24 challenging and very costly. In the simulation case sets used by Kusano et al. (2) and Scanlon et al. (3), the
25 distributions of shoulder width and lane markings crossed prior to road departure were tabulated for U.S.
26 road departure crashes, which are shown in FIGURE 1. Nearly one-third of the vehicles departed roadways
27 without ever crossing any lane markings. In addition, over one-half of the vehicles departed a roadway with
28 no shoulder or minimal shoulder width (<0.3 m).

29
30 FIGURE 1 The proportion of drivers that crossed lane markings and various shoulder width
31 magnitudes prior to their road departure. These values were taken from previous studies (2; 3) that
32 generated these values from a collection of road departures from NASS/CDS 2012.
33 The objective of this study was to quantify the influence of these roadway characteristics on the
34 effectiveness of LDW and LDP within the U.S. vehicle fleet. This study used previously developed models
35 (2; 3) designed to estimate the number of crashes and seriously injured drivers prevented by LDW and LDP
 Scanlon, Kusano, Gabler

1 in order to re-simulate these crashes with hypothetical improvements to roadway infrastructure.

2 METHOD
3 The modeling strategy for this study can be found in FIGURE 2. First, a set of nationally representative
4 road departure crashes was extracted from the 2012 National Automotive Sampling System –
5 Crashworthiness Data System (NASS/CDS) database (24). Second, three simulation case sets with varying
6 roadway infrastructure specifications were formed from this compilation of crashes, including the crashes
7 1) as they actually occurred, 2) with lane markings always present, 3) with an expanded shoulder width (3.6
8 m, i.e. the maximum highway lane width in the U.S.), and 4) with lane markings always present and an
9 expanded shoulder width. Third, each of the crashes within each simulation case set were simulated as if
10 the vehicle was (1) not equipped with LDW or LDP, (2) equipped with LDW, or (3) equipped with LDP.
11 Fourth, the probability that the driver experienced a crash and/or serious injury was estimated for each
12 trajectory. Fifth, the proportion of crashes and seriously injured drivers that could be prevented if the vehicle
13 was equipped with LDW or LDP was computed for each of the simulation case sets, and then compared
14 between roadway infrastructure groups.
15

16
17 FIGURE 2 Overall modeling strategy used in this study to determine the influence of lane markings and
18 shoulder width on the potential effectiveness of LDW and LDP.
19 Data Source
20 NASS/CDS 2012 was used for this study to formulate a simulation case set. This nationally representative
21 database is comprised of 4,000-5,000 crashes collected annually that occurred at various locations
22 throughout the United States. For a crash to be included in this database, at least one vehicle involved must
23 have been towed away from the scene due to damage. This database is ideal for this study for a number of
24 reasons. First, information is recorded about the occupant, vehicle, and environment at the time of the crash.
25 Second, investigators prepare scene diagrams depicting the crash, and take pictures of the crash location.
26 Third, detailed medical records are provided for each occupant involved. In addition, each case is assigned
27 a national weighting factor, which when applied, allows for nationally representative estimated to be made
28 from the dataset.
29 This study specifically analyzed single vehicle road departure crashes that occurred as a result of an initial
30 drift out of lane event. The process for identifying these cases can be seen in TABLE 1. First, coded
31 variables within the NASS/CDS database were initially used to identify drift out of lane road departure
32 crashes, and exclude crashes that may have occurred due to control loss or an evasive maneuver attempt
 Scanlon, Kusano, Gabler

1 (e.g. avoiding an animal in the roadway). Second, for each crash, a manual inspection of evidence (e.g.
2 scene diagram, scene photographs, crash narrative) prepared by the crash investigator was performed to
3 verify that the case was properly coded as a road departure crash. Third, cases were additionally excluded
4 for having a disproportionately high sample weight (cases representative of >5,000 crashes nationally
5 greatly skew the data (25)), departures at T-intersections, and road departures that occurred over multiple
6 departure sides (e.g. the driver over-corrected and exited the opposite side of the road).
7 TABLE 1 Method for compiling drift out of lane road departure crashes from the 2012 NASS/CDS
8 database. Cases were eliminated in the sequence depicted.
Weighted
Group Count
Count
All Crashes in CDS 2012 3,581 1,996,016
Drift out of Lane Departures 629 293,937
Valid Departure after Manual Inspection 556 271,810
Weight > 5,000 5 91,577
Exclusions for Valid
End Departures 8 1,767
Departures
Multi-side Departures 65 30,804
Final Compilation of Crashes 478 147,662
9
10 Formulation of a Simulation Case Set
11 Determining Parameters of Interest
12 The NASS/CDS database contains extensive coded information about the events leading up to crash.
13 However, the coded database lacks some vital information required for adequately reconstructing road
14 departure crashes. For this study, these data were determined through manual review of event records and
15 previously developed statistical models.
16 The initial travel lane prior to the initial lane departure was not provided within the coded NASS/CDS
17 database. Crashes occurring on two-lane undivided highways did not require manual review. On multilane
18 roads, the scene narrative and scene diagram were used to determine this initial travel lane.
19 The presence of lane markings and shoulder width were additionally not coded in NASS/CDS. However,
20 both of these parameters can be determined using scene photographs. The presence of lane markings was
21 examined at the location of initial lane departure. No evaluation of lane marking clarity was made. Shoulder
22 width was estimated on the roadside of the lane departure. Because exact measurements were not possible
23 from the scene photographs, shoulder widths were categorized into four groupings, including 1) no shoulder
24 (<0.3 m), 2) 0.3 m to 1.0 m, 3) 1.0 m to 3.6 m, and 4) >3.6 m.
25 Some information from the crash, including departure angle, departure velocity, and road radius of
26 curvature, had to be determined using previously developed statistical models. In methods described in
27 Kusano et al. (2), the National Cooperative Highway Research Program (NCHRP) 17-22 database was used
28 to formulate statistical models that could predict these three missing parameters from information about the
29 crash. This database contains 890 reconstructed road departure crashes that occurred from 1997 to 2004,
30 and contains the information provided in the NASS/CDS database along with supplemental data. Statistical
31 model development was done in two steps. First, one-way ANOVA were used to determine factors that are
32 significantly correlated with each of the departure conditions. Second, multivariate ANOVAs with
33 combinations of these factors were fit to the data to maximize the adjusted-R2.
34 Simulation Case Set Replications
35 Because some of the conditions of the crash had to be estimated using visual approximations or statistical
36 models, each case was replicated multiple times to account for all of the possible scenarios. All of these
37 replications were then given an equal probability of occurring. For any given case, up to five different
38 variables could have multiple values, including departure angle, departure velocity, road radius of curvature,
39 shoulder width, and driver reaction time. Departure angle, departure velocity, and road radius of curvature
40 (if the road was curved) were each represented as three values that were representative of the 17th, 50th, and
41 83rd percentile of three equally partitioned regions of a normal probability distribution function. Cases with
 Scanlon, Kusano, Gabler

1 shoulder widths between 0.3 m to 1.0 m and 1.0 m to 3.6 m were simulated twice using the maximum
2 possible values of each category. For instance, a case width a shoulder width of 0.3 m to 1.0 m would be
3 simulated with both a shoulder width of 0.3 m and 1.0 m. Driver reaction times were always simulated
4 using two separate values. In summary, a crash that took place on a curved road with a shoulder width
5 between 0.3 to 1.0 m would be simulated under 108 different conditions (3 radius of curvatures × 3
6 departure angles × 3 departure velocities × 2 shoulder widths × 2 driver reaction times).
7 Vehicle Trajectory Simulation
8 Three different types of vehicle trajectory simulations were run in this study, including (1) driver only, (2)
9 driver with an LDW-equipped vehicle, and (3) driver with an LDP-equipped vehicle. LDW and LDP were
10 modeled as becoming activated at the instance that the leading wheel touched the lane line. LDW location
11 relative to the lane marking varies by make and model (26), but is generally within +/- 0.5 m of the lane
12 markings. For simulations without LDW/LDP and/or without visible lane, driver steering was assumed to
13 begin after some reaction time following initial road departure. For simulations where LDW becomes
14 activated, driver steering was assumed to begin after a prescribed reaction time. For simulations where LDP
15 becomes activated, automated vehicle steering was assumed to take place followed by driver steering after
16 some reaction time.
17 Two equally probable driver reaction times, 0.38 s and 1.36 s, were simulated in this study. These values
18 were selected from the observed upper and lower bounds of driver reaction times following a LDW during
19 previous driving simulation studies (12; 27).
20 Vehicle simulations were performed using Carsim vehicle simulation software (28). All numerical
21 integration within each simulation was performed using the 4th order Runge-Kutta method, and a time step
22 of 0.001 s. The roadway lane width was assumed to be 3.48 m wide for divided highways and 3.64 m wide
23 for undivided highways. These lane widths were determined from average values taken from the NCHRP
24 17-22 database.
25 A driver recovery model previously developed by Volvo, Ford, and UMTRI through the Advanced Crash
26 Avoidance Technologies Program (VFU ACAT) was used to simulate driver reaction (4). This model
27 modulates steering using a proportional controller that considers the current heading of the vehicle with
28 respect to the road edge, and the required yaw rate required to maintain the vehicle within the roadway. It
29 was developed using experimental studies conducted in Ford’s VIRTTEX driver simulator. The current
30 study simulates all trajectories using a 2000 Ford Taurus Carsim model, which matches the specifications
31 of Ford’s simulator.
32 A general depiction of the function of the LDP system can be seen in FIGURE 3. At the instance the leading
33 wheel touches the lane marking an LDW is delivered and automated vehicle steering begins. The system is
34 assumed to manipulate steering wheel angular displacement at a rate that linearly ramps up lateral
35 acceleration (rate = 0.2 g/s) to a maximum value of 0.1 g. After some reaction time, driver steering would
36 begin.

37
 Scanlon, Kusano, Gabler

1 FIGURE 3 Depiction of the LDP vehicle trajectory modulation used in this study.
2 Estimating Crash or Serious Injury Reduction
3 After trajectories had been generated for each of the cases within the simulation cases sets, the potential
4 effectiveness of LDW and LDP can be estimated for each of the roadway infrastructure scenarios. This
5 process is done in three main steps. First, the probability that a crash took place was calculated using the
6 off-road vehicle trajectory. Second, given the probability that a crash occurred, the probability of a seriously
7 injured driver was calculated using statistical models that consider the departure conditions of the crash.
8 Third, the potential crash and serious injury reduction, or effectiveness, is calculated for each of the roadway
9 infrastructure scenarios. These methods were previously developed and described by Kusano et al. (2).
10 Previously developed models that calculate the probability of a crash using the off-road trajectory of the
11 vehicle were used in this study, i.e. it was assumed to be impossible for a vehicle to have experienced a
12 crash if it never departs the roadway. These models were developed using data from the NCHRP 17-22
13 dataset. In general, two separate factors were assumed to dictate risk of experiencing a crash: (1) distance
14 travelled laterally away from the roadway and (2) total off-road travel distance. The equation for calculating
15 the probability of a crash can be found in Equation (1), where P[Crashi,j] is the probability that the vehicle
16 experienced a crash during trajectory j of case i, Li,j,k is the total simulated distance travelled within some
17 predefined off-road zone k, Ck is the total number of crashes from NCHRP 17-22 within zone k, and γk is
18 the total trajectory length in zone k from the NCHRP 17-22 data.

K
Ck Li,j,k
𝑃[𝐶𝑟𝑎𝑠ℎ𝑖,𝑗 ] = 1 − ∏ exp (− ) (1)
γk
k=1

19 Probability of a seriously injured driver was then calculated for each trajectory. A seriously injured driver
20 was defined to have sustained a Maximum Abbreviated Injury Score (MAIS) of 3 or greater (3+) according
21 to the AIS version 1998 injury coding (29). Logistic regression models were then formulated using the
22 NCHRP 17-22 database. Given that a crash occurred, the models uses seat belt use and departure velocity
23 to predict probability that the driver sustained an MAIS 3+ injury. Probability of an injury for the given
24 trajectory, P[Injuryi,j], was then calculated using equation (2), where P[Injuryi,j|Crashi,j] is the probability
25 that the driver sustained a serious injury given that a crash occurred.

𝑃[𝐼𝑛𝑗𝑢𝑟𝑦𝑖,𝑗 ] = P[𝐼𝑛𝑗𝑢𝑟𝑦𝑖,𝑗 |𝐶𝑟𝑎𝑠ℎ𝑖,𝑗 ]P[𝐶𝑟𝑎𝑠ℎ𝑖,𝑗 ] (2)

26 In order to calculate the potential crash and safety benefits of LDW and LDP for each roadway infrastructure
27 scenario, the number of crashes (Ncrashes) and seriously injured drivers (NMAIS3+) had to be determined for
28 vehicles equipped with LDW or LDP. Because nationally representative weightings and injury information
29 are provided for each of the cases, the number of crashes and seriously injured drivers are known from the
30 simulation case set. The number of crashes and injured drivers with LDW or LDP could then be calculated
31 using Equation (3) or Equation (4), respectively, where wi,j is the ratio of the case weight to the number of
32 representative simulations.
33
# Cases # Simulations
P[𝐶𝑟𝑎𝑠ℎ𝑖,𝑗 ]
LDW/LDP
Ncrash with LDW/LDP = ∑ ∑ wi,j (3)
i=1 j=1
P[𝐶𝑟𝑎𝑠ℎ𝑖,𝑗 ]
No LDW/LDP

# Cases # Simulations
P[𝐼𝑛𝑗𝑢𝑟𝑦𝑖,𝑗 ]
LDW/LDP
Ninjured with LDW/LDP = ∑ ∑ wi,j (4)
i=1 j=1
P[𝐼𝑛𝑗𝑢𝑟𝑦𝑖,𝑗 ]
No LDW/LDP
 Scanlon, Kusano, Gabler

1 The effectiveness of these systems was then calculated as the potential reduction in crashes and seriously
2 injured drivers using Equation (5).

Nwithout LDW/LDP − Nwith LDW/LDP

ϵ=
Nwithout LDW/LDP
(5)

3 RESULTS
4 A total of 478 crashes from the 2012 NASS/CDS database were used in this study, which were
5 representative of 147,662 crashes nationally. Over 20% of these crashes resulted in a seriously injured
6 driver. The simulation case sets with no roadway infrastructure improvements and added lane markings
7 were both comprised of 20,118 unique simulations. The simulation case sets with only expanded shoulders
8 and with added lane markings + expanded shoulder widths were each composed of 13,290 simulations. A
9 total of 66,816 unique simulations were performed in this study.
10 As shown in FIGURE 1, over one-half of the drivers departed roadways without crossing any shoulder
11 (<0.3 m). Approximately one-third of the drivers departed roads which did not have lane markings that
12 could have potentially activated an LDW/LDP system. A very low percentage of crashes (3.9%) took place
13 on roadways where the driver crossed a shoulder width greater than 3.6 m, which is the maximum highway
14 lane width in the U.S. A cross tabulation of lane markings by shoulder width is additionally shown in
15 TABLE 2.
16 TABLE 2 A cross-tabulation of the presence of lane markings at the location of vehicle departure versus
17 the roadway shoulder width. All presented values were weighted using NASS case weightings.
Shoulder Width (m)
Measure
<0.3 m 0.3 m - 1.0 m 1.0 m - 3.6 m > 3.6 m Total
No 12,090 8.2% 3,414 2.3% 1,947 1.3% 11 0.0% 17,463 11.8%
Lane Markings Present
Yes 73,640 49.9% 31,523 21.3% 19,248 13.0% 5,789 3.9% 130,199 88.2%
Total 85,730 58.1% 34,937 23.7% 21,195 14.4% 5,800 3.9% 147,662 100.0%
18
19 TABLE 3 gives a complete summary of the calculated crash and safety benefits for each simulation case
20 set. With no roadway infrastructure improvements, LDW was found to potentially prevent 28.4% of road
21 departure crashes and 20.7% of seriously injured drivers, and LDP could prevent 32.1% of crashes and
22 27.8% of seriously injured drivers. The results indicate that if all of the vehicles in the dataset had departed
23 roadways with lane markings, LDW could prevent an additional 3.8% of crashes (32.2%) and 6.4% of
24 seriously injured drivers (27.1%), and LDP could prevent an additional 4.3% of crashes (36.4%) and 3.4%
25 of seriously injured drivers (31.2%). If these roadways had expanded shoulder widths (3.6 m), LDW could
26 prevent an additional 21.8% of crashes (50.2%) and 23.4% of seriously injured drivers (44.1%), and LDP
27 could prevent an additional 21.9% of crashes (54.0%) and 20.1% of seriously injured drivers (47.9%). If
28 all of the roadways had lane markings and expanded shoulder widths (3.6 m), LDW could prevent an
29 additional 43.2% of crashes (71.6%) and 39.0% of seriously injured drivers (59.7%), and LDP could prevent
30 an additional 45.6% of crashes (77.7%) and 37.6% of seriously injured drivers (65.4%).
31 TABLE 3 The potential crash and safety benefits of LDW and LDP for each roadway infrastructure
32 scenario. The Effectiveness measure represents the percent of crashes/injuries prevented.

Effectiveness Effectiveness Effectiveness Effectiveness

Values (%) Values (%) Values (%) Values (%)
Added Lane Markings +
Baseline Infrastructure Added Lane Markings Expanded Shoulder Width
Measure Expanded Shoulder Width

Crashes

No LDW/LDP 147,662 --- 147,662 --- 147,662 --- 147,662 ---

 Scanlon, Kusano, Gabler

with LDW 105,657 28.4% 100,080 32.2% 73,540 50.2% 41,963 71.6%

With LDP 100,261 32.1% 93,920 36.4% 67,909 54.0% 32,946 77.7%

Injuries (MAIS3+)

No LDW/LDP 30,167 --- 30,167 --- 30,167 --- 30,167 ---

with LDW 23,871 20.7% 22,121 27.1% 16,948 44.1% 12,232 59.7%

with LDP 21,722 27.8% 20,862 31.2% 15,800 47.9% 10,488 65.4%

1
2 In order to examine the influence of shoulder width, the potential effectiveness of LDW and LDP by
3 simulated shoulder width was plotted for the baseline infrastructure dataset. For shoulder widths of 1.0 m,
4 an estimated 11-14% of crashes could be prevented with LDW/LDP. However, the results indicate that 59-
5 67% of the crashes with shoulder widths of 3.6 m could have been prevented by LDW/LDP.

6
7 FIGURE 4 The potential effectiveness of LDW and LDP for various simulated shoulder widths using the
8 baseline infrastructure dataset. The dataset consisted of 16,920 simulations, because it was restricted to
9 only crashes that took place without any adjacent lanes being crossed (e.g. a right side departure while
10 travelling in the leftmost lane).
11 DISCUSSION AND LIMITATIONS
12 The results provide compelling evidence for the importance of roadway infrastructure on the potential
13 effectiveness of LDW and LDP. Given the current roadway infrastructure, the simulations indicate that
14 LDW/LDP could prevent 28%-32% of single vehicle, drift out of lane road departure crashes, and 21%-
15 28% of seriously injured drivers from these crashes. If lane markings were present on all roadways, these
16 systems were estimated to prevent 32%-36% of crashes and 27%-31% of all seriously injured drivers. If
17 the shoulders were expanded to 3.6 m, LDW/LDP were estimated to prevent 50%-54% of all crashes and
18 44%-48% of all seriously injured drivers. If lane markings were present and the shoulders were expanded
19 to 3.6 m, LDW/LDP were estimated to prevent 72%-78% of all crashes and 60%-65% of all seriously
20 injured drivers.
21 The results suggest that modifying all roadway infrastructure to include lane markings would have a modest
22 impact on LDW/LDP effectiveness with respect to expanding roadway shoulder width. This is attributed to
23 the fact that a large number of the road departure crashes took place on roadways with narrow shoulders
24 (i.e. over one-half of crashes took place on roadways with negligible shoulder width, <0.3 m). Additionally,
25 roadways without lane markings tended to have narrower shoulders. For example, 57% of roadways had a
26 shoulder width less than 0.3 m when lane markings were present, while 69% of roadways had a shoulder
27 width less than 0.3 m when no lane markings were present.
28 The infrastructure improvements proposed in this study would be very costly. However, the results from
29 these simulations highlight how dependent LDW and LDP effectiveness are on these roadway infrastructure
30 features. The performance of these active safety systems currently rely heavily on the ability to not only
31 detect an imminent lane departure, but also steer the vehicle back onto the roadway. Although providing
 Scanlon, Kusano, Gabler

1 wider shoulders that yield additional time and space to steer the vehicle back onto the roadway is a less
2 easily solvable task, the development of road edge detection technologies may help to improve the
3 performance of these systems. Additionally, modifications to the LDW/LDP timing algorithm and LDP
4 vehicle trajectory manipulation may help to reduce the likelihood that the vehicle ever departs the roadway.
5 There are some important limitations to the findings of the current study. First, LDW/LDP were assumed
6 to become activated at the instance the leading wheel touches the lane markings. However, past work has
7 shown that current production vehicles equipped with LDW deliver warnings both before and after initial
8 lane departure depending on the vehicle make and model (14; 26). Additionally, these systems can be
9 deactivated by drivers, or may not become activated if some speed threshold is not reached. Second, the
10 LDP system implemented is a simplified representation of how actual systems work in the fleet. The
11 potential benefits of an LDP system are expected to be dependent on the magnitude of the system’s input,
12 and the mechanism used to alter vehicle trajectory (e.g. steering angle control or selective braking of the
13 vehicle’s wheels). Third, these systems were assumed to become activated on all roadways with lane
14 markings. However, poor visibility and marking clarity will vary by individual weather and road conditions.
15 Fourth, traffic in adjacent lanes and objects on the roadway were not considered in this study. Information
16 about traffic density is not provided in NASS/CDS, so the approach used provides a best case scenario.
17 Fifth, this study’s benefits estimate methodology assumed that the improved roadway infrastructure features
18 would not affect driver recovery in the no LDW/LDP scenario. This assumption may lead to an
19 overestimation of the effectiveness of the performance of LDW/LDP in these scenarios.
20 CONCLUSIONS
21 This study has investigated the influence of roadway infrastructure upon LDW and LDP effectiveness. On
22 current roadways, LDW and LDP were estimated to prevent up to 32% of drift out of lane road departure
23 U.S. crashes and up to 28% of all seriously injured drivers. The results suggest that LDW/LDP could prevent
24 up 36% of crashes and 31% of seriously injured drivers if all roadway had lane markings present. If
25 roadways were to have expanded shoulders, the results indicate that LDW/LDP could prevent up to 54%
26 of all crashes and 48% of seriously injured drivers. If these roadways had lane markings and expanded
27 shoulders, these systems could prevent up to 78% of crashes and 65% of seriously injured drivers. Although
28 the proposed infrastructure modifications would be costly, the results highlight the importance of lane
29 markings and adequate shoulder width on LDW/LDP system effectiveness. Although providing vehicles
30 with adequate shoulder width is a less practical countermeasure, a road edge detection algorithm for
31 LDW/LDP may eventually help to solve the issue of absent lane markings.
32 ACKNOWLEDGEMENTS
33 The authors would like to acknowledge the Toyota Collaborative Safety Research Center (CSRC) and
34 Toyota Motor Corporation for funding this study. Our special thanks to Rini Sherony, Katsuhiko Iwazaki,
35 and Hiroyuki Takahashi of Toyota for sharing their technical insights and expertise throughout the project.
36 We also gratefully acknowledge Kristin Dunford and Kaitlyn Wheeler for their assistance in examining
37 NASS/CDS case documentation. The authors would like to thank Nicholas Johnson for his assistance
38 reviewing the statistical models for predicting missing variables.
39 REFERENCES
40 [1] Kusano, K. D., and H. C. Gabler. Comprehensive target populations for current active safety systems
41 using national crash databases. Traffic injury prevention, Vol. 15, No. 7, 2013, pp. 753-761.
42 [2] Kusano, K. D., T. I. Gorman, R. Sherony, and H. C. Gabler. Potential Occupant Injury Reduction in the
43 US Vehicle Fleet for Lane Departure Warning–Equipped Vehicles in Single-Vehicle Crashes. Traffic injury
44 prevention, Vol. 15, No. sup1, 2014, pp. S157-S164.
45 [3] Scanlon, J. M., K. D. Kusano, R. Sherony, and H. C. Gabler. Potential Safety Benefits of Lane Departure
46 Warning and Prevention Systems in the U.S. Vehicle Fleet.In Proceedings of the 24th International
47 Enhanced Safety of Vehicles Conference, Gothenburg, Sweden, Paper Number 15-0080, 2015.
48 [4] Gordon, T., H. Sardar, D. Blower, M. Ljung Aust, Z. Bareket, M. Barnes, A. Blankespoor, I. Isaksson-
49 Hellman, J. Ivarsson, B. Juhas, K. Nobukawa, and H. Theander. Advanced Crash Avoidance Technologies
50 (ACAT) Program - Final Report of the Volvo-Ford-UMTRI Project: Safety Impact Methodology for Lane
 Scanlon, Kusano, Gabler

1 Departure Warning - Method Development and Estimation of Benefits.In, Department of Transportation,

2 2010.
3 [5] Tanaka, S., T. Mochida, M. Aga, and J. Tajima. Benefit Estimation of a Lane Departure Warning System
4 using ASSTREET. SAE International Journal of Passenger Cars-Electronic and Electrical Systems, Vol.
5 5, No. 2012-01-0289, 2012, pp. 133-145.
6 [6] Teraoka, E. Y. M., S. Tanaka, and T. Mochida. Benefit Estimation Method for Lane Departure Warning
7 Systems in the American Traffic Environment.In, SAE Technical Paper, 2014.
8 [7] Najm, W., and A. Burgett. Benefits estimation for selected collision avoidance systems.In MOBILITY
9 FOR EVERYONE. 4TH WORLD CONGRESS ON INTELLIGENT TRANSPORT SYSTEMS, 21-24
10 OCTOBER 1997, BERLIN.(PAPER NO. 1022), 1997.
11 [8] Houser, A., D. Murray, S. Shackelford, R. Kreeb, and T. Dunn. Analysis of benefits and costs of lane
12 departure warning systems for the trucking industry.In, 2009.
13 [9] Pomerleau, D., T. Jochem, C. Thorpe, P. Batavia, D. Pape, J. Hadden, N. McMillan, N. Brown, and J.
14 Everson. Run-off-road collision avoidance using IVHS countermeasures. DOT HS, Vol. 809, No. 170,
15 1999, p. 88.
16 [10] Krueger, S., J. Abele, C. Kerlen, H. Baum, T. Geißler, S. Grawenhoff, J. Schneider, and W. Schulz.
17 Exploratory study on the potential socio-economic impact of the introduction of intelligent safety systems
18 in road vehicles. VDI/VDE Innovation+ Technik GmbH, Institute for Transport Economics at the University
19 of Cologne, Tech. Rep, 2005.
20 [11] Stanley, L. M. Haptic and auditory cues for lane departure warnings.In Proceedings of the Human
21 Factors and Ergonomics Society Annual Meeting, No. 50, Sage Publications, 2006. pp. 2405-2408.
22 [12] Suzuki, K., and H. Jansson. An analysis of driver's steering behaviour during auditory or haptic
23 warnings for the designing of lane departure warning system. JSAE review, Vol. 24, 2003, pp. 65-70.
24 [13] Chen, M., T. Jochem, and D. Pomerleau. AURORA: A vision-based roadway departure warning
25 system.In Intelligent Robots and Systems 95.'Human Robot Interaction and Cooperative Robots',
26 Proceedings. 1995 IEEE/RSJ International Conference on, No. 1, IEEE, 1995. pp. 243-248.
27 [14] Kusano, K. D., and H. C. Gabler. Comparison of expected crash and injury reduction from production
28 forward collision and lane departure warning systems. Traffic injury prevention, Vol. 16, No. sup2, 2015,
29 pp. S109-S114.
30 [15] Kim, S.-Y., and S.-Y. Oh. A driver adaptive lane departure warning system based on image processing
31 and a fuzzy evolutionary technique.In Intelligent Vehicles Symposium, 2003. Proceedings. IEEE, IEEE,
32 2003. pp. 361-365.
33 [16] Rudin-Brown, C., and Y. Ian Noy. Investigation of behavioral adaptation to lane departure warnings.
34 Transportation Research Record: Journal of the Transportation Research Board, No. 1803, 2002, pp. 30-
35 37.
36 [17] Ford Motor Company. 2015 Ford Explorer Owners Manual. 2014.
37 [18] Mercedes-Benz USA. 2016 S-Class Operator's Manual.In, 2015.
38 [19] Volvo Car Corporation. 2016 Volvo XC90 Owner's Manual.In, 2015.
39 [20] Chrysler Group LLC. 2015 Dodge Charger SRT 392 / SRT Hellcat Owner's Manual. 2014.
40 [21] Toyota Motor Corporation. 2015 Lexus NX 200t Owner's Manual. 2014.
41 [22] Wang, R., Y. Xu, and Y. Zhao. A vision-based road edge detection algorithm.In Intelligent Vehicle
42 Symposium, 2002. IEEE, No. 1, IEEE, 2002. pp. 141-147.
43 [23] Zhang, W. LIDAR-based road and road-edge detection.In Intelligent Vehicles Symposium (IV), 2010
44 IEEE, IEEE, 2010. pp. 845-848.
45 [24] NHTSA. National Automotive Sampling System — Crashworthiness Data System: 2013 Analytical
46 User’s Manual.In, U.S. Department of Transportation, 2014.
47 [25] Kononen, D. W., C. A. Flannagan, and S. C. Wang. Identification and validation of a logistic regression
48 model for predicting serious injuries associated with motor vehicle crashes. Accident Analysis &
49 Prevention, Vol. 43, No. 1, 2011, pp. 112-122.
50 [26] Kusano, K. D., H. Gabler, and T. I. Gorman. Fleetwide safety benefits of production forward collision
51 and lane departure warning systems. SAE International Journal of Passenger Cars-Mechanical Systems,
52 Vol. 7, No. 2014-01-0166, 2014, pp. 514-527.
 Scanlon, Kusano, Gabler

1 [27] Kozak, K., J. Pohl, W. Birk, J. Greenberg, B. Artz, M. Blommer, L. Cathey, and R. Curry. Evaluation
2 of Lane Departure Warnings for Drowsy Drivers. In Proceedings of the Human Factors and Ergonomics
3 Society 50th Annual Meetingof Work, 2006.
4 [28] Mechanical Simulation. CarSim.In, 2012.
5 [29] AAAM. The Abbreviated Injury Scale-1990, Update 1998.In, Des Plaines, IL, 1998.
6

View publication stats