Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
C H A P T E R 1 BackgroundThis chapter introduces the research problem and scope and outlines the organization of the full report. Introduction Lane departure crashes make up a significant number of motor vehicle crashes and account for a disproportionate number of fatalities. LeBlanc et al. (2006) estimated that road departure crashes account for 15,000 fatalities per year in the United States. Neuman et al. (2003), using Fatality Analysis Report- ing System (FARS) data, estimated that almost 39% of traffic fatalities were single-vehicle, run-off-road (ROR) crashes. Moreover, Neuman et al. (2003) evaluated FARS data and estimated that 18% of noninterchange, nonintersection fatal crashes were a result of two-vehicle head-on crashes, the majority in nonpassing situations, with 75% occurring on undivided two-lane roadways. The importance of addressing lane departure crashes has been underscored by the American Association of State High- way and Transportation Officials (AASHTO) in its Strategic Highway Safety Plan. One of the main goals of the plan is to keep vehicles on the roadway; another goal is to minimize the adverse consequences of leaving the road. Furthermore, AASHTO identified mitigating ROR crashes as one of its emphasis areas (Neuman et al., 2003). Lane departure is a serious safety concern, yet the relation- ship between the factors that influence whether a vehicle departs its lane in the first place and the series of actions and events that determine the outcome are complex and not under- stood well. Preventing lane departure in the first place depends on understanding where in the sequence of crash events the lane departure could have been prevented and then applying appropriate countermeasures. For instance, a driver might begin to drift toward the roadway edge because of inattention or because he or she approached a curve at a greater speed than that at which the curve could be negotiated. Both could lead to a road departure, but different countermeasures6would be used for each. Edgeline rumble strips may prevent the first, and in-vehicle curve warning systems may prevent the second. Once a vehicle departs its lane, the ability to ameliorate the outcome depends on understanding the factors that influence each subsequent action or event and then applying the appro- priate countermeasures to address these factors when appro- priate. For instance, a vehicle may partially leave the roadway because of various factors. The ability to recover and safely return the vehicle to the travel lane depends on the roadway, environment, and in some cases, vehicle factors, along with the driverâs response. To illustrate the point, in one situation a driver encounters a fully paved shoulder after leaving the roadway and is able to safely correct and return to his or her travel lane. In an otherwise similar situation, in contrast, the driver encounters gravel shoulders and then overcorrects and loses control, resulting in a rollover. In the first case, the pres- ence of paved shoulders was a key factor in determining sub- sequent events and the final outcome of the road departure. In the second case, the presence of loose shoulder material and the overresponse by the driver exacerbated the situation and resulted in a more serious outcome. The ability to under- stand what factors influence whether a crash occurs, as well as what factors result in less serious outcomes, would greatly improve the ability to select and evaluate the effectiveness of appropriate countermeasures. Currently, understanding why lane departures occur is limited in three ways. First, crash data are limited. Crashes are rare and random, and, as a result, safety analyses must depend on small sample sizes. In addition, crash reporting can be inconsistent, so comparison across sites is difficult. Another problem is the timeliness of crash data. Once a counter- measure is implemented, agencies prefer to evaluate the effec- tiveness as soon as possible before investing more resources in the treatment. However, before-and-after crash studies often cannot be completed for several years after installation of the treatment, until representative samples can be obtained
7and regression to the mean avoided. Crash databases also provide limited information. The sequence of events leading to a crash and the surrounding conditions coded into crash databases are provided by an officer either assessing the situation and recording surrounding conditions after the crash has occurred or questioning drivers and/or witnesses. The amount of information about the sequence of events that precedes a crash, as well as the surrounding roadway, environment, vehicle, and driver conditions that end up in the crash record depends on the reporting officer. The accuracy and usefulness of information that is recorded depend on how carefully the officer evaluates the scene, how accurately or truthfully a driver or witness recalls the sequence of events and conditions that led to the crash, and whether drivers or witnesses actually understood what occurred. For instance, a driver who leaves the roadway and encounters a pavement edge drop-off that causes him or her to lose control as he or she attempts to return to the roadway may not even realize that the reason he or she lost control was because his or her rear tire got caught on the pavement edge. Second, the ability to fully understand lane departures is limited because crash databases only record lane departures that result in a collision. In some cases, collisions, particu- larly minor ones, are not even reported. Additionally, the ability to fully understand lane departures and how they can be prevented or ameliorated requires information about which conditions lead to more favorable events and outcomes. Drivers may leave the roadway edge at the same rate on paved shoulders as on unpaved shoulders but are more likely to fully recover on paved shoulders. Because no crash occurred, there is no record of positive outcomes and which factors influenced them. Third, with current data sets, it has only been possible to study why actual lane departure crashes occur and then to attempt to develop countermeasures for these relatively rare events. Drivers are more likely to be involved in a road depar- ture incident in which they leave the roadway and are able to avoid a crash and return to their lane than they are to be involved in an actual crash. For instance, in the Virginia Tech Transportation Institute (VTTI) 100-car study, researchers found 69 crashes, 761 near crashes, and 8,295 incidents (Klauer et al., 2006). In some cases, these incidents are indi- cators of near misses and provide valuable information about why crashes occur and the crash potential of a given situation. In other cases, positive outcomes indicate that some roadway, environmental, or human factor had a signif- icant positive influence in making a safe return to the road- way possible. Existing naturalistic driving studies, and the future on-road study scheduled under the second Strategic Highway Research Program (SHRP 2) with approximately 2,000 instrumentedvehicles, will provide rich and unique databases that can be utilized to derive relationships among incidents, crashes, and human factors; roadway elements; environmental conditions; and vehicle characteristics and thus address the problems pre- sented in the previous paragraphs (TRB, 2007). Because crashes are rare, crash surrogates have been used to better understand crash risk and overcome many of the problems with crash databases. Naturalistic driving studies obtain information on normal driving but can also capture crashes, near crashes, and incidents that may arise. The frequency of incidents and near- crash incidents is typically greater than actual crashes and can provide greater insights related to the circumstances preced- ing the incident, including the driverâs behavior and any envi- ronmental, roadway, or traffic conditions that may have contributed to the incident. Data from incidents and near crashes can therefore be used as crash surrogate measures to further examine crash risk. Using crash surrogates also provides an opportunity to study what happens before and after an incident. The most significant advantage of the naturalistic driving studies is that they provide a firsthand record of the events that precede crashes and incidents. Roadway, environmental, vehicle, and human factors can be extracted directly, rather than through secondhand information from police records and crash data- bases, to develop relationships among the factors that influ- ence road departure crash risk. Improved data about actual events leading to both road departure crashes and noncrash incidents will be extremely valuable in developing a better understanding of what negative factors lead to crashes and near misses, as well as what factors result in more positive subsequent events and outcomes. Understanding the reasons why crashes do not occur yields as much useful information as evaluating why they do occur. In both cases, factors that cause a vehicle to initially leave the roadway and the relation- ship among road, environment, vehicle, and human factors and subsequent events and outcomes can be studied. Dingus et al. (2006) reported that the analysis of near crashes from the VTTI naturalistic driving study (NDS) has been valuable, as it demonstrates drivers successfully performing evasive maneuvers. Scope of Research The main goal of the research discussed in this report was to identify relevant research questions for addressing lane departures and to determine whether these questions can be addressed using data that is expected to result from SHRP 2âs full-scale NDS. For the present project, lane departure crash surrogates were also identified. The research addressed rural lane departures, with a focus on rural, two-lane, paved roadways.
8To accomplish this agenda, the researchers performed the following tasks: ⢠Driver, roadway, environmental, and vehicle factors expected to contribute to lane departure crashes (summa- rized in Chapter 2) were identified through a review of available literature and through the teamâs expertise. ⢠Data from the VTTI 100-car study and the UMTRI road departure crash warning (RDCW) field operation test (FOT) NDS were evaluated to determine whether the data ele- ments identified could be extracted. The methodology and protocol for extracting those data elements were outlined. The methodology and protocol were described so that this information could be used to extract data from the full- scale NDS (summarized in Appendices A and B). ⢠The accuracy, frequency, and resolution of data collection that would be necessary to address lane departure research questions were determined and summarized (described in Chapter 4). Data elements were also prioritized because resource limitations in the full-scale study will constrain data collection. ⢠A framework for extracting data elements from existing naturalistic studies that can be used for the full-scale study was developed. Appendices A and B describe the protocols, methods, and variable descriptions. Available documenta- tion of the SHRP 2 Safety Projects S03 and S05 work was reviewed to determine what data sensors would be avail- able and what data elements are expected to be available in the full-scale study. The accuracy, frequency, and resolu- tion of data that are expected to be available to answer lane departure questions were evaluated. The team identified limitations and provided feedback to SHRP 2, as described in Chapter 4. ⢠Because data were limited, crash surrogates could be not be evaluated. Chapter 5 summarizes information about lane departure crash surrogates and develops a hierarchy of lane departure crash surrogates that can be used in the full-scale NDS. Existing NDS data were also evaluated to determine starting points for setting triggers to identify lane departure events. ⢠Several analytical approaches that can be used to answer lane departure research questions were developed. Lane departure and normal driving data were identified in the UMTRI RDCW FOT NDS database, and four approaches (data mining, calculation of odds ratio, logistic regression, and a time series analysis) were used to conduct an initial analysis of the data. A description of each approach is pre- sented in Chapter 6. The focus was rural, two-lane, paved roadways. ⢠Input was also provided to researchers for SHRP 2 Safety Project S02, Integration of Analysis Methods and Develop- ment of Analysis Plan. The team collaborated regularlywith the Safety Project S02 team and provided input to research questions. The research presented in this report builds on a Phase I report (Hallmark et al., 2008), relevant background informa- tion from which is included here. Most of the information in this report, however, does not depend on the reader having reviewed the Phase I report. Organization of This Report The remainder of this report is organized as follows: ⢠Chapter 2 provides the results of a literature review con- ducted to identify driver, roadway, environmental, and vehicle factors that have been shown to have some correla- tion to lane departure crashes. Factors identified include horizontal and vertical curvature, roadway cross section, driveway density, illumination, weather, presence of rum- ble strips, roadway delineation and signing, pavement edge drop-off, vehicle type, speeding, influence of alcohol or drugs, driver age, and distraction. Additionally, research questions are identified that may likely be answered using data from the full-scale study or that cannot be answered because of data limitations. The research questions addressed in the scope of this research are also identified. ⢠Chapter 3 summarizes the various data sets used in the research. A description of common data terms is also provided. ⢠Chapter 4 identifies data elements that are expected to be necessary to answer lane departure research questions based on a survey of available literature and the teamâs expertise regarding lane departure issues. The accuracy, frequency, and resolution of each data element are determined and described. Additionally, the availability of the data in the UMTRI and VTTI databases is reviewed and the limita- tions described. The chapter also reviews the available doc- umentation of the SHRP 2 Safety Projects S03 and S05 work. The accuracy, frequency, and resolution of data that are expected to be available to answer lane departure questions in the full-scale study are evaluated. The chapter identifies limitations and provides feedback to SHRP 2, as described in Chapter 5. Data elements are also prioritized because resource limitations in the full-scale study will constrain data collection. ⢠Chapter 5 describes lane departure crash surrogates. Litera- ture on crash surrogates is summarized, and a methodolog- ical approach for selecting and applying crash surrogates is outlined. Existing naturalistic driving study data are also evaluated to determine starting points for setting triggers to identify lane departure events. A discussion of ways to partition normal driving data is also evaluated using exist-
9ing data. Lateral offset is compared for several driving situ- ations. Differences are noted between driving on a tangent and on left- and right-hand curves, between night and day- time driving, and between individual drivers. This chapter also provides some guidance on stratifying normal driving by relevant variables. ⢠Chapter 6 describes four analytical approaches that can be used to evaluate naturalistic driving study data and answer lane departure research questions. Lane departures and normal driving cases have been identified in the UMTRI RDCW FOT NDS data, and four approaches (data mining, calculation of odds ratio, logistic regression, and a time series analysis) have been used to conduct an initial analy- sis of the data. A description of each approach is presented in Chapter 6. The data used, a description of the model,results, sample size, and implications for the full-scale study are also discussed. The focus is on rural, two-lane roadways. Because data were limited during the research, the analysis is exploratory, to determine whether the approach is appro- priate for the full-scale study. ⢠Chapter 7 provides a summary of the entire project. ⢠Appendices A and B describe the protocols, methods, and variable descriptions used to extract data from the UMTRI and VTTI naturalistic driving study data sets. The method used to extract the data provides a framework that can be used by other researchers in working with the full-scale study. Data were extracted manually, which consumed a large amount of resources. How lane departures were identified within the UMTRI data set is also discussed in Chapter 5.