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.
Lane departures are involved in a substantial number of motor vehicle crashes and account for a considerable number of fatalities. Single-vehicle, run-off-road crashes account for almost 39% of traffic fatalities. Two-vehicle head-on crashes account for 18% of noninterchange, nonintersection fatal crashes, with 75% occurring on un- divided two-lane roadways. Addressing lane departure crashes is thus a major safety goal in the United States. Lane departures represent a serious safety concern, but the relationships between fac- tors that influence whether a vehicle departs its lane and the actions and events that determine the outcome are complex and not well understood. The focus of the second Strategic Highway Research Program (SHRP 2) safety research plan is a large field study of naturalistic driving behavior and performance using a comprehensive, state-of-the- art instrumentation package installed in the vehicles of volunteer participants. The SHRP 2 naturalistic driving study (NDS) is intended to support a comprehensive safety assessment of how driver behavior and performance interact with roadway, environ- mental, and vehicular factors and the influence of these factors and their interactions on collision risk, especially the risk of lane departure and intersection collisions. SHRP 2âs safety research plan will produce a database of naturalistic driving behav- ior data under Safety Project S06, Technical Coordination and Quality Control, and Safety Project S07, In-Vehicle Driving Behavior Field Study. Vehicle data will be avail- able from the naturalistic driving data. Environmental data may also be extracted from the outside video views from the vehicle instrumentation system. Safety Projects S04A, Roadway Information Database Developer, Technical Coordination, and Quality Assurance for Mobile Data Collection, and S04B, Mobile Data Collection, will produce a database of roadway characteristics that can be linked to the naturalistic driving data- base to support safety analysis. Some roadway data will be obtained from existing data sets belonging to state and local roadway agencies in the areas where naturalistic driv- ing study data are collected. Select roadway data elements will also be collected using mobile data units. Some roadway information may also be obtained from the vehicle instrumentationâs outside video views. The resulting databases will help researchers bet- ter understand how combinations of driver behavior and roadway, environmental, and vehicle factors lead to different outcomes. Specifically, some of the data will be used in a full-scale evaluation of lane departures in SHRP 2 Safety Project S08, Analysis of the SHRP 2 Naturalistic Driving Study Data. In preparation for the SHRP 2 NDS, the primary goals of the research discussed in this report are to identify lane departure research questions that can be answered using data collected in the field study, identify data needs to address the questions, and to demonstrate analytical methods that can be used to answer those research questions. Executive Summary1
2To accomplish these goals, the following tasks were performed: 1. Identify important research questions related to lane departures. 2. Review and extract data from existing naturalistic driving studies to assess the types of data that may be available in these types of studies. 3. Review information about the types of data that will be available from the SHRP 2 naturalistic driving study. 4. Identify which of the identified lane departure research questions (Task 1) are likely to be feasible for the full-scale study. 5. Identify types of crash surrogates that may be appropriate for use in answering lane departure research questions. 6. Use existing naturalistic driving data sets to explore methods for analyzing SHRP 2 field study data so as to answer the identified questions (Task 4). Although lane departures can occur on any roadway type, this report addresses rural lane departures, with a focus on rural, two-lane, paved roadways. Identification of Lane Departure Research Questions and Necessary Factors One of the main goals of the research was to identify a set of research questions that could be answered using the SHRP 2 NDS and the roadway characteristics databases and that would be useful in determining why drivers leave the roadway and which fac- tors result in different outcomes. The identification of feasible research questions is pre- sented in Chapter 2. The team also identified questions that are not likely to be feasible because of data limitations. Research questions identified as being feasible for the SHRP 2 full-scale study of lane departures include the following: ⢠What environmental, roadway, driver, or vehicle factors influence whether a vehicle departs its lane? ⢠What environmental, roadway, driver, or vehicle factors influence lane departure outcome? ⢠What is the impact of lane departure countermeasures on lane departure frequency and outcome? ⢠What is the relationship between lane departure crash surrogates and crashes? Research questions involving the following factors are not likely to be feasible with the data resulting from the SHRP 2 field study: ⢠Driverâs alcohol or drug use; ⢠Pavement surface friction; ⢠Pavement edge drop-off; or ⢠Quantitative measures of rain, snow, or ice on the road. Identification of Data Necessary to Answer Research Questions Another goal of the research was to determine what data would be necessary to answer the identified lane departure research questions. A comprehensive literature review was conducted to identify driver, roadway, environmental, and vehicle factors that have been shown to have some correlation to lane departure crashes. Factors identified include horizontal and vertical curvature, roadway cross section, driveway density, illu- mination, weather, presence of rumble strips, roadway delineation and signing, pave-
3ment edge drop-off, vehicle type, speeding, influence of alcohol or drugs, driver age, and distraction. The necessary accuracy and resolution for the identified factors were also determined. A summary of the data elements is provided in Chapter 4. Once relevant factors were identified, the team reviewed existing naturalistic driving study and roadway data to determine whether it was feasible to obtain each data element identified. This exercise provided insight as to whether data elements were likely to be avail- able in the SHRP 2 field study and how feasible it would be to extract elements that were not readily available. The team obtained a number of events from field operational tests conducted by the University of Michigan Transportation Research Institute (UMTRI) for a road departure curve warning system. The events contained instances where the drivers left their lane, as well as normal driving data on rural roadways. Raw data from their instru- mentation system included vehicle variables (e.g., vehicle location, forward speed, forward acceleration, yaw, pitch, lateral acceleration) that were provided at 10 Hz and forward images that were provided at 2 Hz. Roadway and crash data were also obtained for the UMTRI study area from the Michigan Department of Transportation (MDOT). The team also received 33 crash or near-crash lane departure events from the VTTI 100-car naturalistic driving study for rural roadways. A reduced data set rather than raw data was provided for each event for most variables. A video clip showing views outside the vehicle was also provided. Both the UMTRI and the VTTI data were examined to determine the feasibility of extracting relevant driver, vehicle, environmental, and roadway factors. The availability of the data in the UMTRI and VTTI databases were reviewed and the limitations described. Chapter 3 summarizes the various data sets used in the research. A description of com- mon data terms is also provided. 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. Data were extracted manually, which consumed a large amount of resources. The method used to extract the data provides a framework that can be used by other researchers in working with the SHRP 2 naturalistic driving study data. The accuracy, frequency, and resolution of data elements that were likely to be avail- able in the SHRP 2 naturalistic driving study were also identified through a review of avail- able documentation about the instrumented vehicle data acquisition system (Safety Project S05, Design of the In-Vehicle Driving Behavior and Crash Risk Study) and a review of preliminary roadway data elements identified in Safety Project S03 (Roadway Measure- ment System Evaluation), as discussed in Chapter 4. Data elements were also prioritized because resource limitations in the SHRP field study will constrain data collection. Information about which data elements were necessary to address lane departure research questions and limitations expected in the SHRP 2 field was used to provide input to the proposed instrumented vehicle data acquisition system (Safety Project S05) and to provide feedback to identification of initial data elements in Safety Project S03. The information will also help guide identification and prioritization of roadway data collection in Safety Projects S04A and S04B. Methodological Approach to Selecting Lane Departure Crash Surrogates Lane departure crashes are a key measure of road safety. However, naturalistic driving studies, even the fully deployed SHRP 2 field study, will have limited cases of lane depar- ture crashes. The naturalistic driving study will capture crashes, near crashes, and incidents, as well as normal driving. The frequency of incidents and near-crash events is typically greater than the frequency of crashes; incidents and near-crash events may be used as crash surrogates. Using surrogates will also provide an opportunity to study what hap- pens preceding and following a lane departure event. The most significant advantage of naturalistic driving studies is that they provide a firsthand record of the events that pre- cede crashes and incidents.
4Chapter 5 discusses potential lane departure crash surrogates that can be obtained from naturalistic driving study data. Literature on crash surrogates with an emphasis on lane departure crash surrogates is summarized, and a methodological approach for selecting and applying lane departure crash surrogates is outlined. Existing naturalistic driving study data were also evaluated to determine starting points for setting triggers to identify lane departure events. The team reduced lane depar- ture incidents in the UMTRI data set for data on rural, two-lane roads resulting in 22 right-side and 51 left-side lane departures. Data for which no incident had occurred, which represented normal driving data, were also extracted. The reduced data was used to assess which variables and thresholds may be the most useful in setting triggers to identify lane departure events in the full-scale field study. Data for several kinematic vehicle variables that may signal a lane departure (lateral speed, yaw rate, side accelera- tion, forward acceleration, roll rate, and pitch rate) were identified. Values for the kine- matic variables were compared for normal driving against left- and right-side lane departures using a Wilcoxon Rank Sum Test to determine whether the normal driving data were statistically different from lane departure events. Although the distributions for most variables were determined to be different at the 95% level of significance, a significant amount of overlap exists. This indicates the difficulty in setting thresholds low enough to include all incidents but still high enough so that a large amount of nonincident data does not have to be evaluated. Although there were not enough data to determine what threshold values should be set, initial results suggest that for left-side lane departures, roll rate, yaw rate, side acceleration, and side speed are likely to be good candidates to identify events. Results suggest that for right-side lane departures, yaw rate, side acceleration, and lateral speed are good candidates to identify events. The VTTI and UMTRI data were evaluated separately because different data were available for each. The only kinematic variables available for the VTTI data set were for- ward and side acceleration. In addition, no normal driving data were available. As a result, it was difficult to assess which variables could be used to determine when lane departures had occurred using that data set. Ways to partition normal driving data were also evaluated using the UMTRI data. Lateral offset was compared for several driving situations. Differences were noted between driving on a tangent and on left- and right-hand curves, between night and daytime driving, and between individual drivers. Some guidance on stratifying normal driving by relevant variables was also provided. Chapter 5 only offers an approach for selecting and evaluating lane departure crash sur- rogates. With the data available, the team was not able to conduct an analysis to evaluate the relationship between lane departure crash surrogates and crashes. The UMTRI data set provided a large number of lane departures and normal driving data but did not include crashes. The VTTI data set contained both crashes and lane departure events; however, once the data were partitioned by roadway type, the data were insufficient to conduct an evaluation. Additionally, no normal driving data were provided for comparison. Exploration of Analytical Approaches to Answer Lane Departure Research Questions Four analytical approaches were identified that can be used to evaluate the data result- ing from the SHRP 2 field study and thus answer the lane departure research questions as discussed in Chapter 6: 1. Data mining using classification and regression tree analysis; 2. Simple odds ratio and logistic regression;
53. Logistic regression for correlated data that accounts for repeated sampling among observations (e.g., repeated sampling for the same driver, trip); and 4. Time series analysis. Initial analysis of the lane departure and normal driving events extracted from the UMTRI data as discussed in Chapter 5 was conducted using the four approaches. 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; and data seg- mentation methods are also presented. The focus is on rural, two-lane roadways. Because data were limited during the research, the analysis was exploratory to determine whether the approach is appropriate for the SHRP 2 field study. Method 1 (classification and regression tree) and Method 2 (simple odds ratio and logistic regression) evaluated the likelihood of a left- or right-side lane departure. A sample-based data aggregation approach was used in the classification and regression tree analysis, and an event-based data aggregation approach was used for the logistic regression. Although available sample sizes were limited, both methods produced sim- ilar results. Both indicated that curve radius, driver age, and type of shoulder were rel- evant in explaining lane departures. Logistic regression also indicated that both left- and right-side lane departures were more likely to occur at night and were less likely to occur as lane width increased. The model for left-side lane departures indicated that male drivers were more likely than female drivers to be involved in a lane departure, and the model for right-side lane departures indicated that lane departures are more likely on roadway sections with a higher density of lane departure crashes and for drivers who spend more time traveling 10 mph or more over the posted speed limit. The third method expanded on a varied logistic regression approach based on the logistic regression model just described, which may be better suited to the data from the full-scale study. The fourth method, time series analysis, used continuous data to develop a model to predict offset as a function of several vehicle kinematic variables. The method was devel- oped and explained in such a way that it could be adapted to the SHRP 2 field study to include various explanatory variables, including driver behavior. This approach allows information, such as driver distraction in previous time periods, to be incorporated into the model. Each analytical approach has advantages and limitations for the full-scale study, and selecting an appropriate method depends on the specific research questions posed and the resources available to reduce the data. The VTTI data could not be used in the analysis since only a limited number of events were available once data were extracted for only rural, two-lane roadways.