Crash avoidance potential of four passenger vehicle technologies

Accident Analysis and Prevention 43 (2011) 732–740

Contents lists available at ScienceDirect

Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap

Crash avoidance potential of four passenger vehicle technologies Jessica S. Jermakian ∗ Insurance Institute for Highway Safety, 1005 North Glebe Road, Arlington, VA 22201, United States

a r t i c l e

i n f o

Article history: Received 8 March 2010 Received in revised form 12 July 2010 Accepted 19 October 2010 Keywords: Crash risk Active safety Crash avoidance technology GES FARS

a b s t r a c t Objectives: The objective was to update estimates of maximum potential crash reductions in the United States associated with each of four crash avoidance technologies: side view assist, forward collision warning/mitigation, lane departure warning/prevention, and adaptive headlights. Compared with previous estimates (Farmer, 2008), estimates in this study attempted to account for known limitations of current systems. Methods: Crash records were extracted from the 2004–08 files of the National Automotive Sampling System General Estimates System (NASS GES) and the Fatality Analysis Reporting System (FARS). Crash descriptors such as vehicle damage location, road characteristics, time of day, and precrash maneuvers were reviewed to determine whether the information or action provided by each technology potentially could have prevented or mitigated the crash. Results: Of the four crash avoidance technologies, forward collision warning/mitigation had the greatest potential for preventing crashes of any severity; the technology is potentially applicable to 1.2 million crashes in the United States each year, including 66,000 serious and moderate injury crashes and 879 fatal crashes. Lane departure warning/prevention systems appeared relevant to 179,000 crashes per year. Side view assist and adaptive headlights could prevent 395,000 and 142,000 crashes per year, respectively. Lane departure warning/prevention was relevant to the most fatal crashes, up to 7500 fatal crashes per year. A combination of all four current technologies potentially could prevent or mitigate (without double counting) up to 1,866,000 crashes each year, including 149,000 serious and moderate injury crashes and 10,238 fatal crashes. If forward collision warning were extended to detect objects, pedestrians, and bicyclists, it would be relevant to an additional 3868 unique fatal crashes. Conclusions: There is great potential effectiveness for vehicle-based crash avoidance systems. However, it is yet to be determined how drivers will interact with the systems. The actual effectiveness of these systems will not be known until sufficient real-world experience has been gained. © 2010 Elsevier Ltd. All rights reserved.

1. Introduction Improvements in vehicle crashworthiness during the past decades have increased occupant survivability when crashes occur (Farmer and Lund, 2006), and recent automobile designs are focusing on ways to avoid crashes altogether. Automakers are refining advanced technologies designed to prevent or mitigate crashes and introducing them into a growing number of passenger vehicle models. Electronic stability control, an early example of crash avoidance technologies, already has proven highly effective, with an estimated 49% reduction in fatal single-vehicle crash risk and an estimated 20% reduction in fatal multiple-vehicle crash risk for equipped cars and SUVs (Farmer, 2010). More recently, technologies such as forward collision warning, lane departure warning, side view assist, and adaptive headlights have been

∗ Corresponding author. Tel.: +1 703 247 1565; fax: +1 703 247 1587. E-mail address: [email protected] 0001-4575/$ – see front matter © 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2010.10.020

introduced in the passenger vehicle fleet, primarily in luxury vehicles. Several researchers have reported on the number of crashes that may be prevented by crash avoidance technologies. For example, prior to commercial availability of these systems, the National Highway Traffic Safety Administration (NHTSA) predicted these systems would prevent as many as 791,000 rear-end, 90,000 lane change, and 297,000 road departure crashes in the United States each year (NHTSA, 1996). More recently, Pomerleau et al. (1999) predicted warning systems could prevent 336,000 road departure crashes per year. A 2005 study concluded that warning systems could prevent nearly 14,000 lane change and road departure crashes in the European Union (Abele et al., 2005). Kuehn et al. (2009) used insurance collision claims data coupled with human factors research and determined as many as 24,000 rear-end, 2000 lane change, and 3000 road departure crashes could be prevented in Germany if all vehicles were equipped with several crash avoidance technologies. In a previous study by the Insurance Institute for Highway Safety, Farmer (2008) estimated the maximum potential

J.S. Jermakian / Accident Analysis and Prevention 43 (2011) 732–740

of five crash avoidance technologies — side view assist, forward collision warning/mitigation, emergency brake assist, lane departure warning, and adaptive headlights — using crash records from the 2002–06 files of the National Automotive Sampling System General Estimates System (NASS GES) and Fatality Analysis Reporting System (FARS). Crash types that possibly could have been prevented or mitigated by a crash avoidance system were counted as relevant for that technology. The study was meant to address not only current technologies but also systems that may be available in the future. Current technologies may not function in all circumstance because of limitations inherent in their designs. For example, lane departure warning systems use visible lane markers to track vehicle position within the lane. If lane markers are not visible because the roadway is covered by snow, the system will not work. Several systems use sensors that may not reliably detect other vehicle movements in rain, snow, sleet, or fog. The purpose of the present research is to refine the initial maximum potential crash reduction estimates, accounting for limitations of currently or imminently available systems. It is important to note that the crash avoidance landscape is changing rapidly, and limitations of current systems may not be limitations of future systems. In addition, the actual capabilities of the technologies may extend beyond what is currently described by vehicle manufacturers. The current study examined four crash avoidance technologies: side view assist for intentional lane changing, forward collision warning/mitigation, lane departure warning/prevention, and adaptive headlights. Crash reduction estimates for brake assist systems were not included in the updated estimates because rudimentary brake assist technologies have been available on passenger vehicles for many years. As a result, there is large market penetration of these systems, and many vehicles in the most recent crash files already may be equipped. Crashes relevant to more advanced brake assist technologies and newer autonomous braking systems are addressed by forward collision warning estimates. Crash counts are estimates of the maximum potential applicability of current technologies. Although these estimates account for relevant crash types and known limitations of current crash avoidance systems, they do not account for potential reductions in effectiveness due to driver interactions with the systems. That is, the estimates provide the total number of crashes that may be prevented or mitigated if the systems were on all vehicles, were 100% effective in alerting drivers and drivers had the appropriate responses 100% of the time. Actual effects depend on how drivers accept the technologies and respond to the information provided by the systems and/or the actions taken by the systems. Not all drivers will react appropriately. Drivers may be overwhelmed or annoyed by the warnings, particularly if vehicles are equipped with multiple technologies, allowing for potential simultaneous warnings. If drivers find the systems annoying or not useful, they may disable the systems, rendering them ineffective. Braitman et al. (2009) surveyed drivers of vehicles with the four crash avoidance technologies described herein and reported that, despite some annoyance by the warnings, the majority of drivers left the systems turned on most of the time. Reported use was lowest for the system that required drivers to turn it on (lane departure prevention) rather than for systems that automatically turned on by default. Several researchers have reported on the effectiveness of these systems, taking into account driver interaction with the systems, as found in field operational tests, simulator studies, and other experimental studies. Field tests of a prototype of a road departure warning system showed mixed results (Wilson et al., 2007). The system worked well consistently on roads with clear lane markings but only 36% of the time on nonfreeways. Taking this into account, the authors estimated the system could reduce road departure crashes in the United States by 5000 to 41,000 per year. A prototype forward collision warning system was field-tested by 66

733

drivers for 4 weeks each (Najm et al., 2006). Based on the number of near crash scenarios identified, the system was projected to reduce rear-end collision rates by 10%. Sugimoto and Sauer (2005) estimated that a system with automatic braking in addition to a warning could prevent 38% of rear-end collisions, while Coelingh et al. (2007) predicted a 50% reduction. Crash avoidance technologies, particularly forward collision warning systems, show promise in mitigating crash severity even if the crash is not avoided entirely. Schittenhelm (2009) analyzed the repair history of certain Mercedes S-Class vehicles with and without Distronic Plus, an active cruise control system that incorporates brake assist and forward collision warning/mitigation technologies. The vehicles with Distronic Plus showed a 5% reduction in rate of repairs for the front bumper alone, damage in the least severe crashes; a 15% reduction in rate of repairs for the front bumper and cross member, damage in mid-severity crashes; and a 25% reduction in rate of repairs for the front bumper, cross member, and longitudinal member, damage in the most severe crashes. The ultimate effect of crash avoidance technologies also depends on whether they fundamentally alter the driving task or driver behavior. Braitman et al. (2009) found both negative and positive effects of crash avoidance technologies on driver behavior. Nine percent of drivers with side view assist systems said they change lanes more often, a small percentage of drivers with forward collision warning reported they look away from the road more often (5%) or follow the vehicle ahead more closely (2%), and drivers with active headlights reported a greater willingness to drive at night (40%) and to drive faster (18%). However, almost half of the drivers with forward collision warning said they follow the vehicle ahead less closely and, among owners with lane departure warning and/or prevention systems, 54–64% reported using their turn signals more often and 67–71% said they drift from travel lanes less often.

2. Methods Data were extracted from two national crash databases maintained by the National Highway Traffic Safety Administration (NHTSA). NASS GES contains information from annual probability samples of police-reported crashes in the United States. Approximately 57,000 crashes are sampled each year. When each case is weighted by the inverse of its selection probability, the yearly sample is representative of about 6 million crashes nationwide (NHTSA, 2008). FARS is an annual census of crashes that occur on public roads and result in the death of a vehicle occupant or other involved party within 30 days of the crash. All passenger vehicle records in the 2004–08 NASS GES and FARS files (vehicle body types 1–49) were merged with the corresponding crash records. Records in GES were weighted by their case weights to produce national estimates. Crashes in GES with maximum injury severity coded as incapacitating (A) or nonincapacitating (B) were classified as severe or moderate injury crashes. To account for missing data in the crash files, imputed data were used whenever available in the GES files. Imputing data involves replacing missing values with likely values, in a statistically appropriate manner, and then analyses are conducted on the augmented data set. The first step was to assign each crash to one of nine general crash types. Classification was hierarchical, so any crash with characteristics of more than one category was assigned to the earliest category in the following list: changing lanes, angle, front-torear, single-vehicle, head-on, other front-to-front, sideswipe same direction, sideswipe opposite direction, and other. In other words, changing lanes took precedence over all other categories. This categorization implied no meaning other than providing a useful starting point for identifying relevant crashes for each technology.

734

J.S. Jermakian / Accident Analysis and Prevention 43 (2011) 732–740

A lane-changing crash was defined as one where a passenger vehicle driver struck another vehicle while intentionally changing lanes, merging, or turning. This was based on the precrash vehicle movement variables. The MANEUV I variable in GES has codes 1 = going straight, 2 = decelerating in traffic lane, 3 = accelerating in traffic lane, 4 = starting in traffic lane, 5 = stopped in traffic lane, 6 = passing or overtaking another vehicle, 7 = disabled or parked in travel lane, 8 = leaving a parked position, 9 = entering a parked position, 10 = turning right, 11 = turning left, 12 = making U-turn, 13 = backing up, 14 = negotiating a curve, 15 = changing lanes, 16 = merging, 17 = corrective action to a previous critical event, 97 = other. The VEH MAN variable in FARS has codes 1 = going straight, 2 = slowing or stopping in traffic lane, 3 = starting in traffic lane, 4 = stopped in traffic lane, 5 = passing or overtaking another vehicle, 6 = leaving a parked position, 7 = parked, 8 = entering a parked position, 9 = controlled avoidance maneuver, 10, 11, 12 = turning right, 13 = turning left, 14 = making U-turn, 15 = backing up, 16 = changing lanes or merging, 17 = negotiating a curve, 98 = other. In both files, NHTSA defines crash types using a manner of collision variable. The MANCOL I variable in GES has codes 0 = not a collision with a motor vehicle in transport (i.e., single-vehicle), 1 = rear-end, 2 = head-on, 3 = rear-to-rear, 4 = angle, 5 = sideswipe same direction, 6 = sideswipe opposite direction. In addition, the accident type (ACC TYPE) variable specifies how each vehicle was configured in the crash. Codes range from 0 to 98 and allow for a finer breakdown of manner of collision. So, for example, records with manner of collision coded as head-on were classified as headon only if accident type also was specifically coded as head-on (i.e., values of 50–53). Otherwise, records were classified as other front-to-front collisions. The MAN COLL variable in FARS has codes 0 = not a collision with a motor vehicle in transport (i.e., single-vehicle), 1 = front-to-rear (includes rear-end), 2 = front-to-front (includes head-on), 3 = front-to-side same direction, 4 = front-to-side opposite direction, 5 = front-to-side right angle (includes broadside), 6 = front-to-side angle (direction not specified), 7 = sideswipe same direction, 8 = sideswipe opposite direction, 9 = rear-to-side, 10 = rear-to-rear, and 11 = other. After crashes were classified according to the nine general crash types, they were further separated into nonrelevant and potentially relevant crash types for each of the four crash avoidance technologies. For example, a crash caused by a driver who intentionally changes lanes probably would not be prevented by a lane departure warning system, but may be prevented by a side view assist system. Crashes involving more than two vehicles were classified as nonrelevant in every case because of an inability to clearly define the sequence of events. Once relevant crash types were defined for a given technology, the effects on relevant crash counts of the limitations of systems currently on the market were estimated. Crashes that occurred during inclement weather (e.g., rain, snow) were identified as non-relevant for all technologies using cameras, radar, or LIDAR (LIght Detection And Ranging) sensors. These sensors may not be reliable in poor weather conditions because precipitation can either obscure objects from view of cameras or interfere with reflected radar/LIDAR signals. Crashes that occurred during inclement weather were identified using the atmospheric condition (WEATHER, WEATHER1-2) variables in GES and FARS. Crashes that involved preimpact braking were identified using the corrective action (P CRASH3) variable in GES and avoidance maneuver (AVOID) variable in FARS. Crashes that occurred while negotiating a curve were identified using the precrash vehicle movement (MANEUV I, VEH MAN) variables in GES and FARS. Crashes that involved speeding were identified using the speed related (SPEEDREL) and violations charged (VLTN I) variables in GES and driver related factors (DR CF1-4) and violations charged

(VIOLCHG1-3) variables in FARS. Crashes involving loss of control were identified using the precrash critical event (P CRASH2) variable in GES and driver related factors (DR CF1-4) variable in FARS. 3. Results There were 271,956 crash records in the 2004–08 NASS GES files that involved at least one passenger vehicle. When sampling weights were applied, these records represented approximately 29,124,000 crashes nationwide. There were 165,175 fatal crashes in the 2004–08 FARS files that involved at least one passenger vehicle. Thus for the 5-year study period, there was an average of approximately 5,825,000 crashes per year involving passenger vehicles, of which 33,035 were fatal. Average annual counts for each of the nine general crash types are summarized in Table 1. 3.1. Side view assist systems Side view assist systems use cameras or radar sensors to monitor areas to the side of a vehicle and alert the driver of vehicles in the side blind zones. This technology has the potential to prevent certain crashes involving two vehicles traveling in the same direction when one of the vehicles intentionally changes lanes. Beginning with all passenger vehicle crashes that involved intentional lane changes (Table 1), crashes were subsequently deemed nonrelevant if the crash circumstances would not be addressed by side view assist systems, leaving only certain types of angle, front-to-rear, and sideswipe same direction crashes as relevant (Table 2). For the remaining crash types, system limitations might further reduce the number of applicable crashes. Specifically, side view assist systems use sensors that may be unreliable in inclement weather; therefore all crashes occurring in poor weather conditions were considered non-relevant. Taking these limitations into consideration, relevant crashes accounted for 24% of lane-changing crashes, or approximately 395,000 crashes per year (Table 2). Such crashes are characterized by a vehicle approaching from behind, so relatively few involve moderate-to-serious injury (20,000) or death (393). 3.2. Forward collision warning/mitigation systems Forward collision warning/mitigation systems use cameras, radar, or LIDAR sensors to monitor the area in front of a vehicle and alert the driver of a potential collision with a vehicle or object. Some systems require the driver to take action, whereas other systems, if driver action is not taken, may autonomously brake or steer the vehicle to reduce crash severity or avoid a crash altogether. Most current systems use radar or LIDAR for range assessment and therefore can detect only objects with reflective surfaces. Also, due to limitations associated with identifying the nature of objects reflecting radar signals, most current systems are designed primarily to address front-to-rear crashes with leading vehicles in traffic. This technology has the potential to prevent certain crashes involving two vehicles traveling in the same direction, where one passenger vehicle strikes the rear of the vehicle ahead. Therefore, relevant crashes were a portion of all front-to-rear crashes (Table 1). Crashes were classified by considering the following variables: relation to roadway (REL RWY in GES, REL ROAD in FARS), first harmful event (EVENT1 I in GES, HARM EV in FARS), number of vehicles (VEH INVL in GES, VE FORMS in FARS), driver related factors (DR CF1-4 in FARS), critical event (P CRASH2 in GES), and avoidance maneuver (DRMAN AV in GES, AVOID in FARS). Beginning with all front-to-rear passenger vehicle crashes (Table 1), crashes were deemed nonrelevant if the crash circumstances would not be addressed by forward collision/mitigation

J.S. Jermakian / Accident Analysis and Prevention 43 (2011) 732–740

735

Table 1 Average annual crashes involving passenger vehicles during 2004–08a . Crash type

All crashes

Lane changing/merging/turning Angle (without lane changing) Front-to-rear (without lane changing) Single-vehicle (without lane changing) Front-to-front head-on (without lane changing) Front-to-front other (without lane changing) Sideswipe same direction (without lane changing) Sideswipe opposite direction (without lane changing) Other (e.g., rear-to-rear, end-swipe, unknown) a

1,617,000 602,000 1,673,000 1,605,000 39,000 15,000 180,000 90,000 3000 5,825,000

Nonfatal injury crashes (A or B)

Fatal crashes

192,000 90,000 116,000 257,000 14,000 4000 11,000 15,000 <1000 698,000

4458 3187 1815 17,838 2690 601 519 1794 133 33,035

Columns may not sum to total due to rounding.

Table 2 Annual intentional lane-changing crashes relevant to side view assist systemsa . Crash type

All crashes

Nonfatal injury crashes (A or B)

Fatal crashes

All intentional lane-changing crashes Nonrelevant crash circumstances Single-vehicle More than 2 vehicle Front-to-front Angle, different travel directions Front-to-rear, striking Sideswipe opposite direction (Subtotal nonrelevant crashes) Crashes not addressed by current system limitations Angle, same travel direction, inclement Angle, unknown travel direction, inclement Front-to-rear, struck, inclement Sideswipe same direction, inclement Other, inclement (Subtotal – system limitations) Crashes relevant to current systems Angle, same travel direction, clear Angle, unknown travel direction, clear Front-to-rear, struck, clear Sideswipe same direction, clear Other, clear

1,617,000

192,000

4458

163,000 58,000 3000 892,000 58,000 3000 (1,177,000)

34,000 15,000 1000 117,000 3000 <1000 (171,000)

928 489 396 1561 75 590 (4039)

19,000 <1000 6000 19,000 <1000 (44,000)

1000 <1000 <1000 1000 <1000 (2000)

1 1 5 17 3 (26)

157,000 3000 50,000 186,000 <1000

10,000 <1000 4000 5000 <1000

5 10 83 270 25

395,000 24% 7%

20,000 10% 3%

393 9% 1%

Total relevant Percent of intentional lane-changing crashes Percent of annual passenger vehicle crashes a

Columns may not sum to total due to rounding; percentages calculated using unrounded numbers

Table 3A Annual front-rear crashes relevant to forward collision warning systemsa . Crash type

All crashes

Nonfatal injury crashes (A or B)

Fatal crashes

All front-to-rear crashes Nonrelevant crash circumstances Off roadway More than two vehicles Vehicle/road defect Avoidance maneuver Struck by nonpassenger vehicle (Subtotal nonrelevant crashes) Crashes not addressed by current system limitations Front-to-rear, on road, with braking, inclement Front-to-rear, on road, without braking, inclement Front-to-rear, roadside, with braking, inclement Front-to-rear, roadside, without braking, inclement (Subtotal – system limitations) Crashes possibly relevant or relevant to current systems Possibly relevant Front-to-rear, on road, with braking, clear Front-to-rear, roadside, with braking, clear Relevant Front-to-rear, on road, without braking, clear Front-to-rear, roadside, without braking, clear

1,673,000

116,000

1815

1000 260,000 3000 8000 40,000 (312,000)

<1000 34,000 <1000 1000 5000 (41,000)

30 515 4 101 204 (853)

44,000 152,000 <1000 <1000 (196,000)

2000 7000 <1000 <1000 (9000)

7 63 1 13 (84)

142,000 <1000

10,000 <1000

65 7

1,020,000 2000

56,000 <1000

677 130

1,165,000 70% 20%

66,000 57% 9%

879 48% 3%

Total relevant Percent of front-to-rear crashes Percent of annual passenger vehicle crashes a

Columns may not sum to total due to rounding; percentages calculated using unrounded numbers.

736

J.S. Jermakian / Accident Analysis and Prevention 43 (2011) 732–740

Table 3B Annual single-vehicle crashes relevant to forward collision warning systemsa . Crash type

All crashes

Nonfatal injury crashes (A or B)

Fatal crashes

All single-vehicle crashes Nonrelevant crash circumstances Off roadway On road, vehicle/road defect On road, noncollision (Subtotal nonrelevant crashes) Crashes not addressed by current system limitations On road, nonpedestrian/bicyclist On road, pedestrian/cyclist, with braking, inclement On road, pedestrian/cyclist, without braking, inclement Roadside, pedestrian/cyclist/parked vehicle, with braking, inclement Roadside, pedestrian/cyclist/parked vehicle, without braking, inclement Roadside, fixed object, with braking, inclement Roadside, fixed object, without braking, inclement (Subtotal – system limitations) Crashes possibly relevant or relevant to current systems Possibly relevant On road, pedestrian/cyclist, with braking, clear Roadside, pedestrian/cyclist/parked vehicle, with braking, clear Roadside, fixed object, with braking, clear Relevant On road, pedestrian/cyclist, without braking, clear Roadside, pedestrian/cyclist/parked vehicle, without braking, clear Roadside, fixed object, without braking, clear

1,605,000

257,000

17,838

890,000 7000 46,000 (943,000)

149,000 1000 9000 (159,000)

11,242 30 930 (12,202)

311,000 <1000 5000 1000 12,000 3000 41,000 (374,000)

7000 <1000 3000 <1000 1000 <1000 6000 (18,000)

373 36 309 2 16 14 132 (882)

5000 2000 7000

4000 <1000 1000

422 12 111

60,000 89,000 127,000

33,000 9000 33,000

2857 192 1160

289,000 18% 5%

80,000 31% 11%

4754 27% 14%

Total relevant Percent of single-vehicle crashes Percent of annual passenger vehicle crashes a

Columns may not sum to total due to rounding; percentages calculated using unrounded numbers.

systems (Table 3A). Collisions on roadways or immediately adjacent to roadways (e.g., shoulder, roadside) were considered relevant. Collisions occurring farther off roadways were classified as nonrelevant for two reasons: (1) detection of objects may be unreliable given the environment and possibly erratic path of the vehicle, and (2) a driver who has departed the roadway and roadside may not be in sufficient control of the vehicle to respond to warnings. Collisions preceded by the successful avoidance of an object in the vehicle’s path also were classified as nonrelevant. The erratic path of a vehicle in the midst of an avoidance maneuver seems likely to hinder the detection of obstacles. Also, these types of avoidance maneuvers may be prompted by forward collision warnings. Collisions preceded by vehicle mechanical problems or road defects also were classified as nonrelevant. Such problems may make it difficult for a driver to take appropriate action to avoid a crash. If a driver recognizes the threat of collision and initiates braking, then the warning/mitigation system may be unnecessary. For this reason, crashes with evidence of braking by the striking vehicle were classified as possibly relevant. Estimates of relevant crash counts were calculated both with and without evidence of braking. Thus a range of estimates for crashes relevant to forward collision warning/mitigation systems were presented. Relevant crash types accounted for 61–70% of front-to-rear crashes (Table 1), or approximately 1,022,000–1,165,000 crashes per year (Table 3A). Of these, 56,000–66,000 involved nonfatal injuries and 807–879 involved fatal injuries. Some new systems are expected to be effective in preventing or mitigating crashes with roadside objects, pedestrians, and bicyclists. These systems use multiple sensors and advanced algorithms to identify and characterize a broader range of potential hazards. Therefore, some portion of single-vehicle crashes (Table 1) also may be potentially relevant in the near future. Beginning with all singlevehicle passenger vehicle crashes (Table 1), crashes were deemed nonrelevant if the crash circumstances would not be addressed by forward collision/mitigation systems (Table 3B). If roadside object, pedestrian, and bicyclist crashes were included, relevant crash

types accounted for 17–18% of single-vehicle crashes (Table 1), or approximately 275,000–289,000 crashes per year (Table 3B). Of these, 75,000–80,000 involved nonfatal injuries and 4209–4754 involved fatal injuries. Combining Tables 3A–3B, approximately 1,298,000–1,454,000 crashes annually were relevant to forward collision warning/mitigation systems, depending on whether the systems are designed to address objects, pedestrians, and bicyclists. Of these, 130,000–146,000 involved nonfatal injuries and 5016–5633 involved fatal injuries. In approximately 89% of these crashes, the driver did not brake to prevent the crash. 3.3. Lane departure warning/prevention systems Lane departure warning systems use cameras to track vehicle position within the lane, alerting the driver if the vehicle is in danger of inadvertently straying across lane markings. This technology is probably not relevant to the intentional lane-changing, angle, or front-to-rear crashes (Table 1). They are, however, relevant to the other crash types including single-vehicle, head-on, and sideswipe crashes, if the vehicle inadvertently drifted out of the lane. Beginning with each of these crash types (Table 1), crashes were subsequently deemed nonrelevant if the crash circumstances would not be addressed by lane departure warning systems (Tables 4A–4D). Maneuvers that successfully avoided an obstacle in the vehicle’s path were not classified as intentional lane changes, but they were not inadvertent either. Crashes involving such maneuvers were deemed nonrelevant to lane departure warning/prevention technology. Collisions preceded by vehicle mechanical problems or road defects also were classified as nonrelevant. Such vehicle or road problems may make it difficult for a driver to take appropriate action to avoid a crash. Crashes caused by nonpassenger vehicles straying across lane markings were deemed nonrelevant because the focus of the analysis was passenger vehicle warning systems. Finally, single-vehicle runoff-road crashes on interstates were not relevant because most interstates have edge-line rumble strips that alert drivers to unin-

J.S. Jermakian / Accident Analysis and Prevention 43 (2011) 732–740

737

Table 4A Annual single-vehicle crashes relevant to lane departure warning systemsa . Crash type

All crashes

Nonfatal injury crashes (A or B)

Fatal crashes

All single-vehicle crashes Nonrelevant crash circumstances On roadway Run off road, vehicle/road defect Run off road, avoidance maneuver Run off road, loss of control Run off road, on interstate highway (Subtotal nonrelevant crashes) Crashes not addressed by current system limitations Speed limit <40 mph Snow on roadway Other, speeding, inclement Other, without speeding, inclement (Subtotal – system limitations) Crashes possibly relevant or relevant to current systems Possibly relevant Other, speeding, clear Relevant Other, without speeding, clear

1,605,000

257,000

17,838

435,000 83,000 663,000 182,000 12,000 (1,375,000)

57,000 13,000 104,000 32,000 4000 (210,000)

4957 155 2834 742 917 (9605)

117,000 6000 4000 10,000 (136,000)

19,000 <1000 1000 2000 (22,000)

1936 115 343 345 (2739)

21,000

7000

2376

73,000

19,000

3119

94,000 6% 2%

25,000 10% 4%

5494 31% 17%

Total relevant Percent of single-vehicle crashes Percent of annual passenger vehicle crashes a

Columns may not sum to total due to rounding; percentages calculated using unrounded numbers.

tentional lane departures, making lane departure warning systems redundant. Given current system limitations, single-vehicle, head-on, and sideswipe crashes might not be relevant if the speed limit was below 40 mph or there was snow on the roadway. Current lane departure warning systems do not operate under a speed threshold of approximately 40 mph and will not provide reliable warnings if lane markers are absent or obscured by snow on the roadway or by heavy precipitation. A lane departure warning may be too late to be effective if a driver is speeding. For this reason, crashes with evidence of speeding were classified as possibly relevant. Tables 4A–4D include estimates of relevant crash counts both with and without evidence of speeding. Thus each table provides a range of estimates for crashes relevant to lane departure warning/prevention systems. Table 4A provides counts of single-vehicle crashes that were not relevant, possibly relevant, and relevant to lane departure warning systems. Single-vehicle crashes on roadways typically were pedestrian, bicyclist, or animal strikes. Avoidance maneuvers

may have been attempted, but they were unsuccessful. As with successful avoidance maneuvers, such crashes were classified as nonrelevant because the vehicle did not inadvertently stray from the travel lane. Single-vehicle crashes were coded as loss of control if the driver lost control of the vehicle due to environmental conditions, overcorrecting, or reckless driving. These crashes also were considered nonrelevant. Relevant crash types accounted for 4–6% of single-vehicle crashes (Table 1), or approximately 73,000–94,000 crashes per year (Table 4A). Of these, 19,000–25,000 involved nonfatal injuries and 3119–5494 involved fatal injuries. Relevant crash types accounted for 23–27% of head-on crashes (Table 1), or approximately 9000–10,000 crashes per year (Table 4B). Of these, 4000–5000 involved nonfatal injuries and 1077–1234 involved fatal injuries. Relevant crash types accounted for 24–29% of sideswipe same direction crashes (Table 1), or approximately 44,000–52,000 crashes per year (Table 4C). Of these, 3000–4000 involved nonfatal injuries and 157–203 involved fatal injuries.

Table 4B Annual head-on crashes relevant to lane departure warning systemsa . Crash type

All crashes

Nonfatal injury crashes (A or B)

Fatal crashes

All head-on crashes Nonrelevant crash circumstances More than two vehicles Vehicle/road defect Avoidance maneuver Nonpassenger vehicle out of lane (Subtotal nonrelevant crashes) Crashes not addressed by current system limitations Speed limit <40 mph Snow on roadway Other, speeding, inclement Other, without speeding, inclement (Subtotal – system limitations) Crashes possibly relevant or relevant to current systems Possibly relevant Other, speeding, clear Relevant Other, without speeding, clear

39,000

14,000

2690

<1000 1000 <1000 4000 (5000)

<1000 <1000 <1000 1000 (2000)

280 10 592 140 (1022)

17,000 3000 1000 2000 (23,000)

5000 1000 <1000 1000 (7000)

223 36 36 138 (434)

1000

1000

157

9000

4000

1077

11,000 27% <1%

5000 35% 1%

1234 46% 4%

Total relevant Percent of head-on crashes Percent of annual passenger vehicle crashes a

Columns may not sum to total due to rounding; percentages calculated using unrounded numbers.

738

J.S. Jermakian / Accident Analysis and Prevention 43 (2011) 732–740

Table 4C Annual sideswipe same direction crashes relevant to lane departure warning systemsa . Crash type

All crashes

Nonfatal injury crashes (A or B)

Fatal crashes

All sideswipe same direction crashes Nonrelevant crash circumstances More than two vehicles Vehicle/road defect Avoidance maneuver Nonpassenger vehicle out of lane (Subtotal nonrelevant crashes) Crashes not addressed by current system limitations Speed limit <40 mph Snow on roadway Other, speeding, inclement Other, without speeding, inclement (Subtotal – system limitations) Crashes possibly relevant or relevant to current systems Possibly relevant Other, speeding, clear Relevant Other, without speeding, clear

180,000

11,000

519

8000 5000 1000 47,000 (61,000)

1000 1000 <1000 3000 (5000)

118 5 89 19 (230)

49,000 6000 5000 6000 (67,000)

1000 <1000 <1000 <1000 (2000)

55 4 11 15 (85)

8000

1000

46

44,000

3000

157

52,000 29% 1%

4000 33% <1%

203 39% 1%

Total relevant Percent of sideswipe same direction crashes Percent of annual passenger vehicle crashes a

Columns may not sum to total due to rounding; percentages calculated using unrounded numbers.

Relevant crash types accounted for 22–25% of sideswipe opposite direction crashes (Table 1), or approximately 20,000–22,000 crashes per year (Table 4D). Of these, 3000–4000 involved nonfatal injuries and 489–598 involved fatal injuries. Combining Tables 4A–4D, approximately 146,000–179,000 crashes annually were relevant to lane departure warning/prevention systems, depending on how the systems affect crashes when drivers are speeding. Of these, 28,000–37,000 involved nonfatal injuries and 4842–7529 involved fatal injuries. In 64–82% of these crashes, the driver was not speeding at the time of the crash.

3.4. Adaptive headlight systems Adaptive headlights rotate in the direction of steering and are intended to improve visibility on curved roads. Beginning with all passenger vehicle crashes (Table 1), crashes were considered relevant if they were front-to-rear, single-vehicle, or sideswipe same

direction crashes that occurred on curves in darkness or twilight. These crash types accounted for 4% of front-to-rear, single-vehicle, and sideswipe same direction crashes (Table 1), or approximately 142,000 crashes per year (Table 5).

3.5. Combined effect There was some degree of overlap among the relevant crashes listed in Tables 2–5. For example, the front-to-rear crashes on dark curves included in Table 3A also were included as relevant crashes in Table 5. As a final step, crashes were identified that were relevant or possibly relevant to any of the four technologies without redundancies. With 100% effectiveness, a combination of all four current technologies could prevent or mitigate (without double counting) up to 1,866,000 crashes each year, including 149,000 serious and moderate injury crashes and 10,238 fatal crashes (Table 6). Newer generation forward collision warning systems also may address object, pedestrian, and bicyclist crashes, and this would increase

Table 4D Annual sideswipe opposite direction crashes relevant to lane departure warning systemsa . Crash type

All crashes

Nonfatal injury crashes (A or B)

Fatal crashes

All sideswipe opposite direction crashes Nonrelevant crash circumstances More than two vehicles Vehicle/road defect Avoidance maneuver Nonpassenger vehicle out of lane (Subtotal nonrelevant crashes) Crashes not addressed by current system limitations Speed limit <40 mph Snow on roadway Other, speeding, inclement Other, without speeding, inclement (Subtotal – system limitations) Crashes possibly relevant or relevant to current systems Possibly relevant Other, speeding, clear Relevant Other, without speeding, clear

90,000

15,000

1794

3000 3000 6000 12,000 (25,000)

1000 1000 1000 3000 (6000)

344 13 438 64 (859)

32,000 5000 2000 4000 (43,000)

3000 1000 1000 1000 (5000)

144 60 58 75 (338)

3000

1000

108

20,000

3000

489

22,000 25% <1%

4000 27% 1%

598 33% 2%

Total relevant Percent of sideswipe opposite direction crashes Percent of annual passenger vehicle crashes a

Columns may not sum to total due to rounding; percentages calculated using unrounded numbers.

J.S. Jermakian / Accident Analysis and Prevention 43 (2011) 732–740

739

Table 5 Average annual crashes relevant to adaptive headlight systemsa . Crash type

All crashes

Nonfatal injury crashes (A or B)

Fatal crashes

All crashes involving negotiating dark curves Nonrelevant crash circumstances Lane changing/merging/turning Angle (without lane changing) Front-to-front Head-on (without lane changing) Front-to-front other (without lane changing) Sideswipe opposite direction (without lane changing) Other (e.g., rear-to-rear, end-swipe, unknown) (Subtotal nonrelevant crashes) Crashes relevant to current systems Front-to-rear (without lane changing) Single-vehicle (without lane changing) Sideswipe same direction (without lane changing)

157,000

32,000

2833

4000 1000 3000 <1000 7000 <1000 (15,000)

<1000 <1000 1000 <1000 1000 <1000 (3000)

13 41 166 22 102 4 (348)

6000 133,000 3000

<1000 28,000 <1000

18 2452 14

142,000 90% 2%

29,000 91% 4%

2484 88% 8%

Total relevant Percent of crashes involving negotiating dark curves Percent of annual passenger vehicle crashes a

Columns may not sum to total due to rounding; percentages calculated using unrounded numbers.

the number of unique crashes potentially prevented or mitigated to 2,115,000 crashes each year, including 218,000 serious and moderate injury crashes and 14,106 fatal crashes. 4. Discussion Approximately one-third of the 6 million police-reported crashes each year are relevant to at least one of the following four crash avoidance technologies: side view assist, forward collision warning/mitigation, lane departure warning/prevention, and adaptive headlights. A combination of all four current technologies potentially could prevent or mitigate (without double counting) up to 1,866,000 crashes each year, including 149,000 serious and moderate injury crashes and 10,238 fatal crashes. Of the four technologies, the one with the greatest potential to prevent or mitigate crashes of any severity is forward collision warning/mitigation. This potentially could prevent/mitigate up to 1.2 million crashes, or 20% of the 5.8 million police-reported crashes each year. A forward collision warning/mitigation system could prevent/mitigate up to 66,000 nonfatal serious and moderate injury crashes and 879 fatal crashes each year. New forward collision warning systems that are expected to detect objects, pedestrians, and bicycles could have even greater potential for preventing or mitigating serious crashes. The new systems might prevent or mitigate up to an additional 80,000 nonfatal serious and moderate injury crashes and 4754 fatal crashes each year, depending on effectiveness. Another technology that could affect a large number of fatal crashes is lane departure warning/prevention. This technology potentially could prevent/mitigate up to 179,000 crashes per year (3% of all crashes), including 37,000 nonfatal serious and moderate injury crashes and 7529 fatal crashes. The warning system alone is similar to a roadway technology already available on many roadways — raised or grooved rumble strips along lane boundaries. Vibrations and noise from centerline and edgeline rumble

strips reportedly prevent 25–30% of potential run-off-road, headon, and sideswipe opposite direction collisions (Corkle et al., 2001; Outcalt, 2001; Persaud et al., 2004). Because rumble strips and lane departure warning systems perform similar functions — warning a driver of unintentional departures from the travel lane — the effectiveness of rumble strips provides an indication of how drivers may react to lane departure warning technologies. Based on the published 25–30% effectiveness estimates for rumble strips, a more realistic estimate of crashes that may be prevented by lane departure warning systems likely would be 45,000–54,000 per year. The counts listed in Tables 2–6 are the crashes that potentially could be avoided if these technologies worked perfectly and drivers reacted perfectly, but the counts are not true estimates of system effectiveness. Experience with rumble strips may provide a more realistic estimate of the effectiveness of lane departure warning systems; however, similar proxies are not available for the other systems. The success of crash avoidance technologies in preventing crashes depends on several factors, including driver acceptance and use of the technologies. Even if technologies can work perfectly to prevent crashes, their effectiveness will be limited if drivers are unwilling to use them. Drivers may misunderstand or react inappropriately to the alerts. False alarms may cause drivers to mistrust systems and either ignore them or turn them off (Campbell et al., 2007). However, Braitman et al. (2009) found that the majority drivers of vehicles with the four crash avoidance technologies reported leaving the systems turned on most of the time. The authors focused on early adopters driving luxury vehicles and, therefore, results might not be typical of all drivers. On the other hand, drivers with too much faith in the systems may be less observant or drive more aggressively. Research on reflector posts, raised pavement markers, and other roadway markings on curves has reported that drivers sometimes increase their speeds when visibility is improved (Kallberg, 1993; Zador et al., 1987). This could offset the potential benefits of adaptive

Table 6 Annual crashes that potentially could be prevented or mitigated by four technologies given current system limitations. All crashes

Nonfatal injury crashes (A or B)

Fatal crashes

Side view assist Forward collision warning Lane departure warning Adaptive headlights

395,000 1,165,000 179,000 142,000

20,000 66,000 37,000 29,000

393 879 7529 2484

Total unique crashes Percent of annual passenger vehicle crashes

1,866,000 32%

149,000 21%

10,238 31%

740

J.S. Jermakian / Accident Analysis and Prevention 43 (2011) 732–740

headlights, as confirmed by Braitman et al. (2009) who found that drivers using adaptive headlights reported they were more likely to drive at night and at higher speeds. However, the authors also found that the presence of crash avoidance systems may influence self-reported driver behavior for the better including following the vehicle ahead less closely, using turn signals more often, and drifting from the lane less often. LeBlanc et al. (2006) found similar results from field tests, reporting that drivers using lane departure warning systems improved their lane-keeping behavior, traveling near or beyond the lane edge less frequently, and increased their use of turn signals. Driver interaction with these systems is the subject of ongoing field, track test, and simulator studies. Researchers are trying to determine the best way to warn drivers of dangerous situations and assist in correcting errors. However, drivers in experimental settings, even when using their own vehicles on public roads, do not always behave realistically. The true test of the effectiveness of these crash avoidance systems will not be known until sufficient numbers of drivers gain real-world experience. The crash avoidance features in this study currently are available on just a handful of vehicle models, but they are becoming more common. They also are becoming more sophisticated. After flashing an indicator light and/or sounding an alarm to warn the driver of an imminent collision, some forward collision warning systems tighten safety belts and initiate automatic braking. Some lane departure systems actively resist the movement of a vehicle out of its travel lane. These improvements will not add to the list of potentially relevant crashes, but they could increase the percentage of such crashes that actually will be prevented. In addition, many of the limitations of currently available systems may not be limitations of future systems, and this potentially could increase system applicability in circumstances such as inclement weather. The current study has several limitations. Estimates were based on two databases, NASS GES and FARS. These databases contain data on crashes that occurred in the United States therefore crash distributions may not be representative of European or other nonUS populations. Both databases rely on police-reported data that may include some degree of misclassification of key variables that affect the applicability of crashes within a given category. These databases may not contain sufficient detail on vehicle, driver, or environmental conditions to determine the true applicability of crash avoidance technologies. For example, current lane departure warning systems rely on lane markings to determine vehicle position in the lane, but information with sufficient detail to determine lane markings is not included in the NASS GES and FARS data sets. Therefore, the assumption that lane markings are present on roads with speed limits of 40 mph and higher was made to determine applicability of lane departure warning systems. In addition, the databases do not provide sufficient information to differentiate between crashes that may be avoided altogether or those whose severity may be mitigated. Warning systems require driver action, but driver impairments such as distraction, alcohol use, medical issues, and drowsiness were not considered in this study. Although warning systems might alert a distracted driver to an impending crash, the reaction of impaired drivers to such systems is not well understood and is beyond the scope of the current study. Finally, the capabilities of crash avoidance technologies vary among systems and, although this study attempted to account for the majority of systems currently available, some systems may have capabilities beyond those described or limitations not well characterized.

Acknowledgments The author gratefully acknowledges Charles M. Farmer, Adrian K. Lund, David S. Zuby, and Anne T. McCartt of the Insurance Institute for Highway Safety for their scientific and editorial contributions to this work. This work was supported by the Insurance Institute for Highway Safety. References Abele, J., Kerlen, C., Krueger, S., Baum, H., Geissler, T., Grawenhoff, S., Schneider, J., Schulz, W., 2005. Exploratory Study on the Potential Socio-economic Impact of the Introduction of Intelligent Safety Systems in Road Vehicles (SeiSS Final Report). VDI/VDE Innovation + Technik GmbH, Teltow, Germany. Braitman, K.A., McCartt, A.T., Zuby, D.S., Singer, J., 2009. Volvo and Infiniti Drivers’ Experiences with Select Crash Avoidance Technologies. Insurance Institute for Highway Safety, Arlington, VA. Campbell, J.L., Richard, C.M., Brown, J.L., McCallum, M., 2007. Crash Warning System Interfaces: Human Factors Insights and Lessons Learned. Report No. DOT HS810-697. National Highway Traffic Safety Administration, Washington, DC. Coelingh, E., Jakobsson, L., Lind, H., Lindman, M., 2007. Collision warning with autobrake—a real-life safety perspective. Paper no. 07-0450. In: Proceedings of the 20th International Technical Conference on the Enhanced Safety of Vehicles. National Highway Traffic Safety Administration, Washington, DC. Corkle, J., Marti, M., Montebello, D., 2001. Synthesis on the Effectiveness of Rumble Strips. Report no. MN/RC-2002-07. Minnesota Department of Transportation, St. Paul, MN. Farmer, C.M., 2010. Effects of Electronic Stability Control on Fatal Crash Risk. Insurance Institute for Highway Safety, Arlington, VA. Farmer, C.M., 2008. Crash Avoidance Potential of Five Vehicle Technologies. Insurance Institute for Highway Safety, Arlington, VA. Farmer, C.M., Lund, A.K., 2006. Trends over time in the risk of driver death: what if vehicle designs had not improved? Traffic Injury Prevention 7 (4), 335–342. Kallberg, V., 1993. Reflector posts—signs of danger? Transportation Research Record 1403, 57–66. Kuehn, M., Hummel, T., Bende, J., 2009. Benefit estimation of advanced driver assistance systems for cars derived from real-life accidents. Paper no. 09-0317. In: Proceedings of the 21st International Technical Conference on the Enhanced Safety of Vehicles. National Highway Traffic Safety Administration, Washington, DC. LeBlanc, D., Sayer, J., Winkler, C., Ervin, R., Bogard, S., Devonshire, J., Mefford, M., Hagan, M., Bareket, Z., Goodsell, R., Gordon, T., 2006. Road Departure Crash Warning System Field Operational Test: Methodology and Results. Report no. UMTRI-2006-9-1. University of Michigan Transportation Research Institute, Ann Arbor. Najm, W.G., Stearns, M.D., Howarth, H., Koopman, J., Hitz, J., 2006. Evaluation of an Automotive Rear-End Collision Avoidance System. Report no. DOT HS-810-569. National Highway Traffic Safety Administration, Washington, DC. National Highway Traffic Safety Administration, 2008. Traffic Safety Facts 2007. Report no. DOT HS-811-002. US Department of Transportation, Washington, DC. National Highway Traffic Safety Administration Benefits Working Group, 1996. Preliminary Assessment of Crash Avoidance Systems Benefits. US Department of Transportation, Washington, DC. Outcalt, W., 2001. Centerline Rumble Strips. Report no. CDOT-DTD-R-2001-8. Colorado Department of Transportation, Denver, CO. Pomerleau, D., Jochem, T., Thorpe, C., Batavia, P., Pape, D., Hadden, J., McMillan, N., Brown, N., Everson, J., 1999. Run-Off-Road Collision Avoidance Using IVHS Countermeasures. Report no. DOT HS-809-170. National Highway Traffic Safety Administration, Washington, DC. Persaud, B.N., Retting, R.A., Lyon, C.A., 2004. Crash reduction following installation of centerline rumble strips on rural two-lane roads. Accident Analysis and Prevention 36 (6), 1073–1079. Schittenhelm, H., 2009. The Vision of Accident Free Driving—How Efficient Are We Actually in Avoiding or Mitigating Longitudinal Real World Accidents. Paper no. 09-0510. In: Proceedings of the 21st International Technical Conference on the Enhanced Safety of Vehicles. National Highway Traffic Safety Administration, Washington, DC. Sugimoto, Y., Sauer, C., 2005. Effectiveness Estimation for Advanced Driver Assistance System and its Application to Collision Mitigation Brake System. Paper no. 05-0148. In: Proceedings of the 19th International Technical Conference on the Enhanced Safety of Vehicles. National Highway Traffic Safety Administration, Washington, DC. Wilson, B.H., Stearns, M.D., Koopmann, J., Yang, C.Y., 2007. Evaluation of a Road-Departure Crash Warning System. Report no. DOT-HS-810-854. National Highway Traffic Safety Administration, Washington, DC. Zador, P.L., Stein, H.S., Wright, P.H., Hall, J.W., 1987. Effects of chevrons, postmounted delineators, and raised pavement markers on driver behavior at roadway curves. Transportation Research Record 1114, 1–10.