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.
A P P E N D I X B Methodology for Extraction of Data Elements from the Virginia Tech Transportation Institute Naturalistic Driving Study Data Set The following describes the variables selected to evaluate lane departure research questions using the VTTI naturalis- tic driving study data set (100-car study). It also describes the methodology to extract those variables when relevant. VTTI had already identified near crashes and crashes as part of the 100-car naturalistic driving study data. These were the only data available; there were no exposure data. Thirty-three near crashes and crashes were provided by VTTI after a data request was made as part of this project. VTTI has since made some data from their 100-car study publically available on a data distribution website (http://forums.vtti.vt.edu/index .php?/files/category/3-100-car-data/). Data on rural areas from this website showed that lane departure was involved. These data from the VTTI website had not been included among the data earlier provided to the research team. The nar- rative and other data descriptions for data from both sources were reviewed. Several cases that had been identified as near crash or crash have been found to be the result of intentional lane changes or merging by the subject or another driver. These instances were not included in the present study, because it focuses on unintentional lane departures. Excluding these cases yielded a total of 29 crashes and near crashes that were used in various analyses. However, the VTTI data consisted of all rural roadway types (e.g., ramp, divided, two lane), so there were very few samples for each particular roadway type. A description of the data sets is provided in Chapter 3. VTTI had already identified the start and end times for each near crash or crash to reduce events. Their convention was used to define the start and end points when continuous data were used. Data from VTTI were already reduced in the sense that crashes and near crashes were identified; the video data were already reduced, providing event narratives, eyeglance infor- mation, and so forth. The data were processed into various databases to facilitate analysis. The VTTI data required very little processing and included data for both divided roadways and two-lane roadways. As a122result, both were used in several of the analyses even though only data for two-lane roadways were extracted from the UMTRI data. Only data on rural roadways were included. Since spatial information was not provided with the VTTI data, additional data could not be extracted from other sources, such as aerial imagery. Continuous data were provided at 10 Hz (0.1 s) for each crash or near crash. The following lists the variables that were available in the VTTI data or was extracted and was expected to be used in the analysis. Vehicle Variables Information was obtained by the VTTI data reductionist, unless otherwise indicated. ⢠Vehicle type: No information was provided about vehicle type. ⢠Acceleration: Forward and side acceleration were provided for each record of data, measured in g-force (g). ⢠Speed: Forward speed in mph. ⢠Available maneuvering room forward and rear: Range (ft) and position (degrees) of obstructions in range of radar. ⢠Brake: Brake status (off/on). ⢠Turn signal state: Status of turn signal (off, right, left). ⢠Vehicle factors: Vehicle factors that may have contributed to event (e.g., tire defect or malfunction, wiper defect or malfunction). Driver Variables Information was obtained by the VTTI data reductionist, unless otherwise indicated. ⢠Age: Driver age. ⢠Gender: Driver gender. ⢠Driver reaction: Driver reaction in response to the event (e.g., steered to left, steered to right).
123⢠Driver behavior: Driver actions that occurred near the event, actions that led to the event, or actions taken to avoid the event (e.g., exceeded speed limit, avoided animal, driving without lights). ⢠Driver impairment: Potential driver factors (e.g., drowsy, angry, drugs). ⢠Driver distraction: Up to three distractions that the driver was engaged in 5 to 6 s prior to the onset of the event (e.g., lost in thought, reading). ⢠Hand on wheel: Describes whether driver had hands on wheel (no hands, left hand, right hand, both hands). ⢠Visual obstructions: Factors that may have interfered with driverâs line of sight (e.g., curve, trees, rain). ⢠Reaction of other drivers: Action or maneuvers by other drivers causing the event or in response to the event. Event Variables Information was obtained by the VTTI data reductionist, unless otherwise indicated. ⢠Event start and end: Time stamp that marks approximate start and end of event. ⢠Event nature: Type of conflict that occurred (e.g., conflict with lead vehicle, conflict with merging vehicle). ⢠Preincident maneuver: The action that the vehicle was engaged in just prior to the event (e.g., going straight, stopped in traffic). ⢠Maneuver judgment: Indication of the legality of maneuver leading to the event as determined by the data reductionist. ⢠Precipitating event: Event that started the sequence leading to the crash or near crash (e.g., subject over right-lane line). ⢠Post maneuver: Vehicle action after avoidance of crash or near crash. ⢠Vehicles: Number of vehicles, type of vehicles, maneuver of other vehicles involved in the event. ⢠Fault: Indication of which driver caused the event. ⢠Vehicle position: Position of surrounding vehicles (e.g., in front and to right). Roadway Variables Information was obtained by the VTTI data reductionist, un- less otherwise indicated. ⢠Infrastructure: Roadway factors that may have affected a driverâs ability to safely navigate the roadway (e.g., roadway alignment, weather). ⢠Lanes: Number of traffic lanes in the direction of travel. ⢠Geometry: Presence of curve or grade (e.g., straight/level, curve/level, straight/grade). ⢠Land use: Land use in area at start of the event (e.g., church, residential).Traffic Variables Information was obtained by the VTTI data reductionist, unless otherwise indicated. ⢠Density: Level of service (A to F) as determined by the data reduction. ⢠Traffic control: Traffic control at start of the event (e.g., stop sign). ⢠Intersection: Position of vehicle relative to intersection or junction at the time of the event (e.g., nonintersection, intersection, driveway). Environmental Variables Information was obtained by the VTTI data reductionist, un- less otherwise indicated. ⢠Roadway surface condition: Roadway surface condition that may have caused a reduced coefficient of friction (e.g., wet, dry, ice). ⢠Lighting: Light condition (dawn, daylight, dusk, dark/lighted, dark/not lighted). ⢠Weather: Ambient weather (clear, cloudy, fog, mist, rain, snow, sleet, smoke). Data Limitations A number of variables that were determined to be necessary to answer the lane departure research questions were not available from any source in the VTTI data. A list of data variables neces- sary to answer the lane departure research questions is summa- rized in Chapter 4, which also outlines the limitation of the VTTI data. In summary, the primary limitations that affected the research teamâs ability to fully answer research questions included the following: ⢠No lane tracking information, such as lane width or vehicle offset, was available that could determine vehicle position relative to the lane. The researchers had to rely on the VTTI data reductionistâs interpretation of whether a vehicle had departed its lane. ⢠Vehicle spatial position (latitude/longitude) was not provided. As a result, the researchers could not overlay the vehicle traces with aerial imagery or other spatial data sets to extract additional information, such as radius of curve. ⢠Forward video resolution made it difficult to determine a number of factors that could be determined in the UMTRI forward video. For instance, in the UMTRI data, it was pos- sible to tell the distance to an object on the side of the road- way and to determine the pavement surface condition.
