Evaluation of Data Needs, Crash Surrogates, and Analysis Methods to Address Lane Departure Research Questions Using Naturalistic Driving Study Data (2011)

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 4 Roadway, Driver, Environmental, and Vehicle Data Needs and Limitations to Address Lane Departures Using Naturalistic Driving Study DataBackground This chapter discusses the roadway, driver, environmental, and vehicle factors that must be included to address the research questions outlined in Chapter 2 (“Research Questions,” p. 17). Factors that are expected to be relevant to lane departure crashes were identified through a review of available literature, as well as through the team’s expertise in lane departure issues. Many factors deal with roadway features and the correlation between roadway countermeasures and lane departures or lane departure crashes. The intent of including research questions related to roadway features is to provide roadway agencies with information about the roadway factors that positively or neg- atively influence the likelihood of a lane departure so that agen- cies can better address safety in roadway design and assess the benefits of various countermeasures, such as rumble strips, flattening or better delineating curves, and mandating paved shoulders on reconstruction and rehabilitation projects. Several environmental factors were also identified as con- tributing to lane departures and lane departure outcome. A number of driver factors are also relevant, such as driver char- acteristics (e.g., age, gender, driving experience), distraction, and the driver’s emotional or physical state (e.g., medical con- dition, alcohol or drug use). Vehicle variables are those that may cause a vehicle to be more or less likely to depart its lane and affect the subsequent outcome, such as vehicle braking sys- tem, center of gravity, and vehicle size. Finally, some additional variables that are not used to evaluate the likelihood of a lane departure or outcome are nonetheless necessary to identify vehicle state (e.g., yaw rate, vehicle position). These variables also need to be included in data streams resulting from the full- scale NDS. Hence, they are included in subsequent discussions about necessary data variables. In order to answer these questions and address the relation- ship between lane departures and roadway, driver, environ- mental, and vehicle factors in the full-scale study, data needs should be identified and limitations in data quality, availabil- ity, or accuracy should be addressed. This chapter identifies27factors that are expected to be relevant, identifies sources and limitations from a review of existing naturalistic driving study data (UMTRI and VTTI), and identifies data limitations that are expected to be relevant to the full-scale study. Documentation describing the data sources in the full-scale in-vehicle naturalistic driving study was also reviewed. The expected data elements from mobile mapping (SHRP 2 Safety Project S03, Roadway Measurement System Evaluation) and the in-vehicle instrumentation (Safety Project S05, Design of the In-Vehicle Driving Behavior and Crash Risk Study) rele- vant to lane departures were identified. The availability of the data in the full-scale driving study is also commented on, and limitations are identified. The data sets used to evaluate the feasibility of extracting necessary data elements to answer the stated lane departure research questions are described in Chapter 3. A number of variables were reduced by the team from the data sets described in this chapter. A detailed description of how variables were extracted is provided in Appendices A and B. The next section, “Review of Roadway, Environmental, and Vehicle Data Elements Available in Existing Naturalistic Driving Study Data” (p. 28), summarizes the review of exist- ing data sources to determine whether the necessary data ele- ments could be extracted. This section describes the minimum roadway, driver, environmental, and vehicle data elements nec- essary to answer the research questions. The expected accuracy and resolution requirements are also discussed. Additionally, the availability of the data in the existing data sets and the lim- itations in extracting these data are discussed. An indication of the accuracy and resolution that would be desirable is also provided. The team first reviewed the various data sets that are cur- rently available, as described in the previous sections, and com- mented on their adequacy for answering the lane departure research questions. Another section below, “Review of Planned Data Collec- tion for Full In-Vehicle Naturalistic Driving Study” (p. 39),

28discusses the review of SHRP 2 Safety Projects S03 and S05 documents that describe the data collection systems for the pro- posed full-scale naturalistic driving study. The research team’s understanding of the relevant data collection sensors and tech- niques and the expected accuracy and frequency of data collec- tion are summarized. The data elements were compared with the requirements set out in the next section, “Review of Road- way, Environmental, and Vehicle Data Elements Available in Existing Naturalistic Driving Study Data.” The adequacy and limitations of the methods, data accuracy, and data collection frequency for answering lane departure research questions are discussed. The following summarizes information for road- way, driver, environmental, and vehicle data elements. It should be noted that the review of the full-scale data col- lection methods was based on the Safety Project S03 and S05 documents that were available to the research team as of Sep- tember 2009. The review was also based on the team’s under- standing of the different sensors/methods. The data review was completed before the draft version of this report was provided to SHRP 2 in September. The team has reviewed any informa- tion that has become available during the review period for this report, and as of January 2010, the team has had no additional information that changes the findings presented here. The naturalistic driving study data from UMTRI and VTTI were used to evaluate the variables that may be the most useful in setting triggers to identify lane departure events and to assess what thresholds may be used. Data were reduced as described in Appendices A and B. The UMTRI data resulted in a number of encroachments, but no conflicts or crashes. Only data for rural, paved, two-lane roadways were included. The VTTI data included near crashes and crashes, but no encroachments. Additionally, variables were not consistent between the two data sets. As a result, the two data sets were evaluated separately, as discussed in the fol- lowing sections. Review of Roadway, Environmental, and Vehicle Data Elements Available in Existing Naturalistic Driving Study Data The ability to answer the research questions depends on obtaining the appropriate data about driver, roadway, environ- mental, and vehicle factors that will probably affect the likeli- hood of a lane departure, as well as on obtaining data that are needed to determine crash surrogate thresholds and vehicle position. Data at the appropriate resolution are also necessary to develop measures of exposure. Data from the existing data sets described in Chapter 3, including naturalistic driving study data from UMTRI and VTTI, were reviewed to determine whether necessary dataelements could be extracted. This section describes the min- imum roadway, driver, environmental, and vehicle data ele- ments necessary to answer the research questions. The accuracy and resolution requirements needed are also discussed. The availability of the data in the existing data sets and the limita- tions in extracting these data are discussed. An indication of the accuracy and resolution that would be desirable is also provided. Exposure factors are also included. Each data variable has a list of possible sources. For instance, some types of roadway data could be obtained from aerial imagery, mobile mapping, state databases, or even the forward video from the naturalistic driving study data acquisition sys- tem. In the course of this research, the team often compared information from one source to another as a check. The team encourages researchers who will use the data from the full-scale naturalistic study and other data sets to do the same. Identification of Necessary Variables At a minimum, factors that should be included in causal rela- tionships are those that have already been identified in other studies. A comprehensive literature review was conducted, and a list of potential variables that affect the likelihood and sever- ity of lane departure crashes is reported in Chapter 2 (“Rele- vant Data Elements Identified in Existing Literature,” p. 10). These factors were reviewed with the research questions in mind, and a list of driver, roadway, environmental, and vehi- cle factors that were determined to be important in addressing lane departures were summarized based on this information and the expertise of the research team. In addition to factors that are expected to positively or neg- atively affect the likelihood of a lane departure crash, some other information will also be necessary, such as vehicle posi- tion and vehicle kinematics. Vehicle kinematics is necessary to identify triggers that can be used to flag lane departure events in the full-scale study. Resolution and accuracy were determined on the basis of the team’s experience, common accuracies for the metric, or expert opinions. For instance, superelevation is typically from 2% to 12%. As a result, an accuracy of at least ±0.5% seems log- ical. The desired accuracy of the lane tracking system was spec- ified as 0.1 m. The lane position tracking system is critical for addressing lane departure questions. Several experts were questioned about the level of risk of different lane departure events. The experts unanimously agreed that even one tire leaving the paved roadway surface onto a grass, gravel, or mixed-surface shoulder constitutes a highly dan- gerous situation. As a result, the lane position tracking sys- tem should be accurate and reliable enough to determine when one or more tires have departed the roadway surface. The average tire width is 6 in. (0.15 m), so a desirable accu- racy of ±0.1 to 0.15 m was specified.

29Vehicle Factors Needed to Answer Lane Departure Research Questions The following section summarizes vehicle factors necessary to address lane departure research questions, indicates potential sources in the existing data sets, suggests accuracy and fre- quency needs, and includes comments about accuracy and availability in the existing data sets. Data Element: Vehicle Spatial Position (Latitude, Longitude) • Need: Establishes vehicle position so that data can be linked among spatial databases and spatial relationship between subject vehicle and features established. • Potential source for data element: GPS. • Desired accuracy: Standard GPS accuracy is approxi- mately 2 to 5 m. Accuracy of 5 m is sufficient to locate vehi- cle data to a roadway, but higher accuracy is necessary to determine where a vehicle is at a particular point in time relative to roadway features collected from other sources. • Desired frequency: 10 Hz. • Comments on extracting data from existing data sets: UMTRI was uncertain about the accuracy of the GPS system in their RDCW system but estimated the accuracy to be between 2 and 3 m (from e-mail correspondence). When vehicle traces were overlaid with aerial imagery and corre- lated with forward video, the spatial location of the vehicle appeared quite accurate. GPS data were not provided with the VTTI data set. Data Element: Distance Between Vehicle and Roadside Objects Outside objects are generally those that a vehicle may strike, such as a utility pole or other vehicle, once they leave their original lane of travel. • Need: Establishes vehicle position relative to objects so that time to collision can be calculated and level of risk assessed. • Potential source for data element: (1) Distance can be determined using spatial location of vehicle and object, or (2) distance to objects can be determined using forward or side radar if the object is within the range of the forward or side radar. • Desired accuracy: ±3.0 ft (0.914 m). If a vehicle were trav- eling at 60 mph (80.67 ft/s) and the nearest strikable fixed object it may hit were within 3 ft, the error in calculating total transfer capability would be 3 ft ÷ 80.67 ft/s = 0.0372 s. For a vehicle traveling 35 mph, the error would be 0.058 s. • Resolution: NA. • Comments on extracting data from existing data sets: Even if objects can be located with a high level of accuracy,vehicle position from GPS accuracy is only likely to be at best ±6.56 ft (2 m), so the ability to correctly measure dis- tance is constrained by limitations in GPS accuracy. With use of radar, accuracy is determined by that of the forward or side radar units. The drawback to this method is that the radar can only indicate that an object is within the range of the radar. It will be necessary to identify the object using forward or side video. Data Element: Vehicle Position Within Its Lane • Need: Lane position may be the best indicator of when a lane departure has occurred. Lane position can also be used to determine the magnitude of the lane departure in terms of the departure angle from the roadway and the amount that the vehicle encroaches onto the shoulder. Both can be used to set thresholds between different levels of crash surrogates. • Potential source for data element: Data can only be obtained from lane position tracking algorithms and associ- ated data streams such as forward video. • Desired accuracy: It is not specifically stated, but it appeared that the accuracy of the lane tracking software in the UMTRI data was 0.328 ft (0.1 m). Since this is less than the width of an average tire (around 6 in.), it is expected that this accu- racy is sufficient. • Resolution: Collection of vehicle position at 10 Hz is ade- quate to establish angle of departure and offset. • Comments on extracting data from existing data sets: Lane position was not provided with the VTTI data set. Lane position in the UMTRI data set was given in terms of a measure of lane width and offset from the lane center at a given point in time. Vehicle width was known and constant among all vehicles. Using these three variables, a vehicle’s position within its lane could be determined as shown in Figures 4.1 and 4.2. The data provided in the UMTRI data set were determined to be adequate to extract data neces- sary to answer the research questions. Lack of some type of lane positioning information would seriously affect the ability to determine when crash surrogate events occurred.Data Element: Longitudinal Acceleration (ax) and Speed (vx) • Need: Magnitude of acceleration (positive or negative) can indicate an evasive action and can be used as a measure to determine thresholds between levels of crash surrogates. Acceleration rates can also be used as indicators of aggres- sive driving. • Potential source for data element: Longitudinal acceler- ation and speed are measured from an accelerometer or are output from an on-board system. These data were provided with both the UMTRI and VTTI data sets. Brake engagement data were also provided, which can be used

30Figure 4.1. Determination of lane departure and amount of shoulder encroachment.Figure 4.2. Vehicle tire path calculated from UMTRI data (shows location and amount of encroachment onto shoulder).to indicate that a vehicle is braking (decelerating) but not to provide the magnitude of braking. • Desired accuracy: 0.1 ft/s2 and 0.1 ft/s (0.03 m/s2 and 0.03 m/s). Acceleration is also frequently expressed in “g ’s.” • Resolution: Data collected at 10 Hz should be sufficient. • Comments on extracting data from existing data sets: None. Data Element: Lateral Acceleration (ay) and Lateral Speed (vy) • Need: Indicate side movement, which can be used to deter- mine when a lane departure has occurred and the severity of the lane departure. Lateral acceleration and speed are also used to determine roll hazard. • Potential source for data element: Lateral acceleration and speed, usually measured from an accelerometer, were avail- able in both the VTTI and UMTRI data sets.• Desired accuracy: 0.1 ft/s2 and 0.1 ft/s (0.03 m/s2 and 0.03 m/s). Acceleration is also frequently expressed in “g ’s.” • Resolution: Data collected at 10 Hz should be sufficient. • Comments on extracting data from existing data sets: None. Data Element: Pitch, Roll, Yaw • Need: Define vehicle rotation around several axes and are used to define levels between crash surrogate thresholds and assess roll hazard. • Potential source for data element: Usually measured from an accelerometer. • Desired accuracy: Unknown. • Resolution: Data collected at 10 Hz intervals should be sufficient. • Comments on extracting data from existing data sets: This data element was available in the UMTRI data set. No limitations were noted. Data Element: Presence and Distance Between Subject Vehicle and Other Vehicles • Need: Establish outcome from lane departure and are used as a measure of level of service. Presence of other vehicles (opposing, vehicles passed) can be used to determine road- way density as an exposure method. • Potential source for data element: Forward or side video, forward or side radar. • Desired accuracy: ±3 ft (0.914 m). • Resolution: Collected as they occur. • Comments on extracting data from existing data sets: Oncoming vehicles and vehicles that were passed or that passed the subject vehicle could be determined from the for- ward video from both UMTRI and VTTI. However, only a subjective measure of distance could be obtained from the forward video, as shown in Figure 4.3 (following closely, following, forward vehicle present but not following, no

31(a) (b) Image source: UMTRI RDCW data set. Figure 4.3. Subjective measurement of vehicle following: (a) subject vehicle is following forward vehicle and one oncoming vehicle is passing subject vehicle; and (b) subject vehicle not considered to be following forward vehicle.forward vehicle); distance could not be determined. Dis- tance to a forward or side vehicle could be determined from the forward or side radar. However, only vehicles within the radar range could be detected. Roadway Factors Needed to Answer Lane Departure Research Questions The following section summarizes roadway factors necessary to address lane departure research questions, indicates potential sources in the existing data sets, suggests accuracy and fre- quency needs, and includes comments about the accuracy and availability in the existing data sets. Several factors that are highly relevant but that will not be available in any of the data sets include the presence, type, and amount of pavement edge drop-off and pavement surface friction. A number of studies have suggested that drop-off can con- tribute to outcome when a driver leaves the edge of the road- way and that pavement edge drop-off–related crashes tend to be more severe than other types of run-off-road crashes (Hallmark et al., 2006). The team provided this as a sugges- tion during discussions about the in-vehicle instrumentation package. However, the only method to collect these data with the in-vehicle instrumentation is to use a camera pointed at the roadway. This would capture presence and type of drop-off, but not its amount. Pavement edge drop-off could be collected with the mobile mapping units, but the amount and presence of drop-off can vary significantly over time. As a result, record- ing drop-off with the mobile mapping units would only be rel- evant for that point in time. Surface friction is also an important factor in run-off-road crashes, particularly on curves. However, none of the vehicleinstrumentation packages can capture this variable. It can and may be collected by the mobile mapping units. However, as with pavement edge drop-off, surface friction can vary signif- icantly over time, especially in climates where winter weather maintenance is frequent. As a result, collection of this variable may only be representative of a particular point in time. Data Element: Lane Width • Need: Independent variable in the statistical analysis. Also needed to establish vehicle position within its lane. • Potential source for data element: When using lane width as an independent variable, data can be obtained from exist- ing roadway data sets or from mobile mapping. Lane width is expected to be collected using the lane position tracking system. Lane width was available with the UMTRI data from the forward lane position tracking system. • Desired accuracy: An accuracy of 0.328 ft (0.1 m) is likely the best that can be achieved with the forward lane position track system. • Resolution: Data collected at 10 Hz should be sufficient. • Comments on extracting data from existing data sets: The ability to measure lane width using the lane position track- ing system is critical for establishing vehicle position within the lane and determining when and by how much a vehicle departs its lane. Lane width was not provided with the VTTI data set and could not be measured using any of the data provided. Data from the UMTRI system were adequate for research needs. Data Element: Roadway and Shoulder Surface Type • Need: Independent variable in statistical analyses. The type of shoulder will also affect potential outcomes for lane departures. • Potential source for data element: Existing roadway data sets or mobile mapping. Roadway and shoulder type could also be determined from forward video. • Desired accuracy: Categorical data (should include asphalt, concrete, gravel, earth). • Resolution: Several times per mile or when characteristics change. • Comments on extracting data from existing data sets: While identification of features was possible with forward imagery from the VTTI and UMTRI data sets, color imagery would enhance the ability to distinguish features. Data Element: Shoulder and Median Width • Need: Independent variable in statistical analyses. Shoulder and median width also affect potential outcomes for lane departures.

32• Potential source for data element: Existing roadway data sets or from mobile mapping. Roadway and median width could also be measured using forward video from UMTRI when distances were calibrated. • Desired accuracy: ±0.5 ft (0.152 m). • Resolution: Several times per mile or when characteristics change. • Comments on extracting data from existing data sets: Shoulder width was calculated in the UMTRI data set using side radar. The shoulder width measurement, however, was inaccurate because it was only measuring whether an object was located within the radar range. Shoulder and median width could be calculated when a distance was calibrated in the forward imagery. Data Element: Number of Lanes, Access Control, and Presence and Type of Median • Need: Establish roadway type. An independent variable in statistical analyses. • Potential source for data element: Mobile mapping, for- ward imagery, aerial imagery, roadway databases. • Desired accuracy: NA. • Resolution: Once per mile or when characteristics change. • Comments on extracting data from existing data sets: Roadway type, number of lanes, and type of access control were provided with both UMTRI and VTTI data. Whether or not the roadway was divided was indicated, but no infor- mation about presence and type of median was included. Roadway characteristics could easily be determined from the aerial imagery as well. Data Element: Curve Length and Radius • Need: Independent variable in statistical analyses. May also be used to assess roll hazard. • Potential source for data element: Mobile mapping or aerial imagery. • Desired accuracy: ±25 ft (7.62 m). • Resolution: Once per curve. • Comments on extracting data from existing data sets: Radius was measured from the lane position tracking system in the UMTRI data, but the data received were inaccurate. Radius and curve length were measured from aerial imagery. The accuracy of this method is not known. Data Element: Curve Superelevation, Lane Cross Slope • Need: Independent variable in statistical analyses. May also be used to assess roll hazard. • Potential source for data element: Mobile mapping is likely the only feasible source.• Desired accuracy: Maximum superelevation for areas with no ice and snow is 12%; for areas with snow and ice the max- imum is 8%. Given these ranges, ideal accuracy is 0.5%, but it is unknown if this accuracy can be practically measured in the field. Under normal circumstances cross slope is 1.5% to 2%. Ideally, it would be necessary to measure this variable at 0.1% accuracy to determine differences, but this may not be practical. • Resolution: Superelevation would need to be measured as it changes along a curve. Cross slope could be collected several times per mile or when characteristics change. • Comments on extracting data from existing data sets: Superelevation and lane cross slope were not available in any data sets used and could not be extracted from other sources. Data Element: Curve Direction from Perspective of Driver (Curve Left or Right) • Need: Independent variable in statistical analyses. Also important for determining the potential outcome of a non- crash lane departure. • Potential source for data element: Needs to be determined for direction of travel. Potential sources are aerial imagery or mobile mapping. A forward video image can also be used to determine direction. • Desired accuracy: NA. • Resolution: Should be indicated once per curve. • Comments on extracting data from existing data sets: None. Data Element: Distance Between Curves • Need: Drivers may negotiate curves differently if they travel for some distance between curves rather than negotiate a series of curves. Also used as an independent variable in statistical analyses. • Potential source for data element: Aerial imagery or mobile mapping. Distance could also be calculated from the vehicle instrumentation. • Desired accuracy: ±25 ft (7.62 m). • Resolution: Once per curve. • Comments on extracting data from existing data sets: Could be determined from either aerial imagery or the UMTRI vehicle traces. Both were adequate. Data Element: Type and Characteristics of Curve Spirals • Need: Independent variable in statistical analyses. • Potential source for data element: Mobile mapping is the only reasonable method that can be used to determine the presence of spirals, along with their characteristics, such as radius and length. • Desired accuracy: ±25 ft (7.62 m).

33• Resolution: Once for each spiral. • Comments on extracting data from existing data sets: The team reviewed aerial imagery but could not detect or mea- sure spirals. No information on spirals was provided with any of the data sets, and this information could not be extracted. Data Element: Amount of Grade (Percent), Length of Grade (ft or m), and Location and Characteristics of the Crown and Crest Vertical Curve • Need: Independent variable in statistical analyses. Also affects lane departure outcome. • Potential source for data element: Existing roadway data- bases and mobile mapping are the best sources. Presence of vertical grade and where a vehicle is relative to a vertical curve can be determined from forward video, as shown in Figure 4.4. An estimate could be determined from topo- graphic maps, but this would be time-consuming.Source: UMTRI RDCW data set. Figure 4.4. Forward imagery indicating that vehicle is currently on an upgrade.• Desired accuracy: 0.5% for grade and ±25 ft (7.62 m) for length. • Resolution: Could be measured each time grade changes and at beginning and ending points on vertical curves. • Comments on extracting data from existing data sets: Grade was not provided in any of the data sets reviewed. As indicated in Figure 4.4, grade could subjectively be deter- mined in the UMTRI data set by viewing the forward imagery. The amount or length of grade could not be determined. Data Element: Signing (Would Include Features Such as Overhead Beacons, Signals, and Other Traffic Control Signs) As a minimum, signs collected should include all traffic con- trol signs (stop and yield), warning signs (e.g., overhead flash- ing beacons, curve warning, curve advisory speed, change in alignment warnings, as shown in Figure 4.5), railroad crossing signs and markings, and school crossing signs and markings.• Need: Independent variable in statistical analyses. • Potential source for data element: Mobile mapping, exist- ing sign inventories, forward video. • Desired accuracy: The general location of the sign or an indication that the sign is present is adequate. For instance,it would be important to know the number and type of chevrons along a curve, but it would not be necessary to know exactly where each is located. It is also assumed that all signs are compliant with National Cooperative Highway Research Program (NCHRP) 350 so that they would not need to be considered as strikable fixed objects when deter- mining the outcome of a lane departure event. A sign located using a standard GPS with accuracy of ±6.6 ft (2 m) would be adequate. • Resolution: As they occur. • Comments on extracting data from existing data sets: The main limitation in the data sets reviewed was that they did not include most of the signing data. The UMTRI data set provided the speed limit and advisory speed when known, but no other sign information was available. A sign’s pres- ence could be detected in most cases in the forward imagery for the UMTRI data set. However, because forward imagery was only provided at 5 Hz (two images per second), in some cases depending on where the sign was relative to the vehi- cle’s position when the image was taken, the lettering on the sign could not be distinguished (especially at night). In most cases, the vehicle was not close enough in one frame, and in the next frame the vehicle had passed the sign. Data Element: Number of Driveway or Other Access Points (Driveways/Mile, Access Points/Mile) • Need: Traffic entering and exiting the traffic stream can impact vehicle operation. This traffic would be included as an independent variable in statistical analyses. • Potential source for data element: Mobile mapping, aer- ial imagery, forward imagery. • Desired accuracy: NA. • Resolution: NA. • Comments on extracting data from existing data sets: Driveway and other access point densities for each vehicle trace in the UMTRI data set were determined by overlaying the traces with aerial imagery. Driveways and access points could be identified from the forward imagery in most cases, but this would be an extremely time-consuming method to extract the data. Data Element: Presence and Type of Edge and Centerline Rumble Strips • Need: Independent variable in statistical analyses. Also needed to establish outcome of lane departure. • Potential source for data element: Roadway databases, mobile mapping, and forward video. • Desired accuracy: NA. • Resolution: Once per mile or when rumble strip starts or ends. • Comments on extracting data from existing data sets: This element would best be determined from mobile mapping,

34(a) (d) (c) Source: FHWA 2007. (b) Figure 4.5. Types of signing to be included: (a) horizontal alignment warning signs; (b) vertical grade warning signs; (c) miscellaneous warning signs; and (d) roadway condition and advance traffic control warning signs.but it also can be determined from the forward video, as shown in Figure 4.6, particularly with color. However, char- acteristics such as width, depth, or skip distance would be difficult or impossible to extract from images.Data Element: Roadway Delineation (Presence of Lane Lines or Other On-Roadway Markings) • Need: Critical for lane position tracking software. Would be included as an independent variable in statistical analyses. • Potential source for data element: An initial estimate could be obtained from mobile mapping. Some states, such asIowa, have good pavement marking inventories. These sources could be used as references. However, pavement markings can wear fairly quickly under adverse conditions, so a method to determine pavement marking condition cur- rent to each driving situation would be necessary. This infor- mation could only come from forward imagery and would need to be a qualitative assessment (i.e., highly visible, visi- ble, obscured, not present). Figure 4.7 shows an example of a subjective measure.• Desired accuracy: Data would include a quantitative esti- mate of visibility of markings. • Resolution: Once per mile or as the situation changes.

35Figure 4.6. Image from DriveCam showing presence and type of rumble strips.(a) (b) (c) Source: UMTRI RDCW data set. Figure 4.7. Subjective measure of lane marking condition using forward imagery: (a) pavement markings indicated as “highly visible”; (b) pavement markings indicated as “visible”; and (c) right pavement markings indicated as “obscured.”• Comments on extracting data from existing data sets: This element needs to be current to the driving situation and can only be extracted from forward imagery. This information could be obtained from the UMTRI data set, but was more difficult with the VTTI data set because of image resolution. Data Element: Location and Type of Roadside Objects • Need: Necessary to determine potential outcome of lane departures. May be included as an independent variable instatistical analyses. (Features such as guardrail along the edge of the roadway may impact driver behavior.) • Potential source for data element: Mobile mapping would be the primary source. Some data can be obtained from aer- ial imagery. The presence of fixed objects can be identified in the forward imagery. Forward and side radar readings can be used to determine presence and distance of objects within range. • Desired accuracy: Since location of fixed objects will be used to determine time to collision or potential outcome of a lane departure, a high level of accuracy is desirable. It is expected that ±3 ft (0.914 m) is sufficient. • Resolution: Roadside objects should be collected when they appear; data should be collected for objects within the clear zone. • Comments on extracting data from existing data sets: Only a limited number of fixed objects, such as trees, could be determined from the aerial imagery (image resolution was approximately 1 to 3 m depending on the area). Pres- ence of fixed objects was also identified in the forward imagery from UMTRI. A rough estimate of distance from the edge of the roadway could also be made, but this was not accurate enough to assess the outcome of lane depar- tures. Forward and side radars can indicate the presence but not the type of objects that are within the radar range. Additionally, only objects within the range of the radars can be identified. Environmental Factors Needed to Answer Lane Departure Research Questions The following section summarizes environmental factors nec- essary to address lane departure research questions, indicates potential sources in the existing data sets, suggests accuracy and frequency needs, and includes comments about the accuracy and availability in the existing data sets. Data Element: Roadway Surface Condition (Weather Related, as Well as Presence of Roadway Irregularities Such as Potholes) • Need: Independent variable in statistical analyses. May also impact potential outcome of lane departure. • Potential source for data element: Forward- or other outward-facing video, status and frequency of wiper blades, outside temperature if available, roadway weather infor- mation system (RWIS) data if archived. • Desired accuracy: Measure is subjective and therefore inapplicable. • Resolution: Collected at 10-min intervals or as conditions change.

36• Comments on extracting data from existing data sets: A subjective measure of roadway surface condition and roadway irregularities could be obtained from both the VTTI and UMTRI forward video. Measures such as pres- ence of water on the roadway can be determined as shown in Figure 4.8. Amount of water on the roadway or presence of ice and vertical elevation differences between lanes and shoulder (i.e., pavement edge drop-off ) cannot be deter- mined with any available data sources.(a) (b) (c) Source: UMTRI RDCW data set. Figure 4.8. Pavement surface condition from forward imagery: (a) snow present but roadway bare; (b) wet but amount of water cannot be determined; and (c) surface irregularities.Data Element: Environmental Conditions Such as Raining, Snowing, Cloudy, and Clear (May Not Correspond to Roadway Surface Condition) • Need: Independent variable in statistical analyses. May affect sight distance and is related to visibility. • Potential source for data element: Forward imagery or archived weather information, ambient temperature probe. • Desired accuracy: Will be a subjective measure. • Resolution: Collected at 10-min intervals or when condi- tions change significantly. • Comments on extracting data from existing data sets: A general assessment of environmental conditions could be obtained from the forward video provided in the UMTRI and VTTI data sets. Even with wiper position known, it was difficult to tell how heavy the rainfall was. Archived weather information can provide general information for an area but cannot tell the exact environmental conditions in the loca- tion of the subject vehicle.Data Element: Ambient Lighting, Includes Presence of Street Lighting • Need: Independent variable in statistical analyses. • Potential source for data element: Sun angle, dawn, dusk, day, and night indicator can be obtained from time stamp data and U.S. Naval Observatory astronomical data. They can also be obtained from light meter and headlamp use. • Desired accuracy: Only subjective measures will be used. • Resolution: Can be recorded as the situation changes (day to dusk) or when significant changes occur during the day because of clouds. The measure can be somewhat generic (dark, dark with continuous lighting, dawn, dusk, daytime clear, or daytime with need for headlamps). • Comments on extracting data from existing data sets: A rel- ative estimate of ambient lighting could be obtained in most cases from the UMTRI and VTTI forward imagery. The lim- itations are that it was difficult during high cloud cover or low visibility to subjectively estimate ambient lighting. Data Element: Visibility • Need: Independent variable in statistical analyses. Serves as a measure of sight distance and can also indicate surface conditions. • Potential source for data element: Forward or other outside imagery is the only reasonable data source for visibility. • Desired accuracy: Subjective variable, so accuracy is irrelevant. • Resolution: Sampling at 10-min intervals would be sufficient. • Comments on extracting data from existing data sets: This element was available from forward imagery in the VTTI and UMTRI data sets; however, in some cases it was difficult to tell whether visibility or image resolution was the problem, as shown in Figure 4.9. The cause of decreased visibility could not be determined. Low visibility is shown in Figure 4.10, but it is unknown if the cause is fog, smoke, or dust.Source: UMTRI RDCW data set. Figure 4.9. Image shows some reduced visibility, but it may be the result of sun angle or image resolution.Exposure Factors Needed to Answer Lane Departure Research Questions The following section summarizes exposure factors necessary to address lane departure research questions, indicates poten-

37Source: UMTRI RDCW data set. Figure 4.10. Low visibility appears to be caused by fog.tial sources in the existing data sets, suggests accuracy and fre- quency needs, and includes comments about the accuracy and availability in the existing data sets. Data Element: Annual Average Daily Traffic • Need: Exposure measure. • Potential source for data element: Roadway databases; most states have archived in some form. • Desired accuracy: Most current year available. • Resolution: NA. • Comments on extracting data from existing data sets: This element could be determined from the crash data when no other source was available. Data Element: Time Driving Into Trip • Need: Exposure measure. • Potential source for data element: Vehicle data stream. • Desired accuracy: ±1 s. • Resolution: Is expected to be available in at least 1-s intervals. • Comments on extracting data from existing data sets: NA. Data Element: Amount of Driving on Different Roadway Types Under Different Environmental Conditions (Roadway Type, Rural vs. Urban, Dry vs. Snow, Snow vs. Ice, Ice vs. Wet) • Need: Exposure measure. • Potential source for data element: Vehicle data stream. • Desired accuracy: ±1 s. • Resolution: Is expected to be available at least 1-s intervals. • Comments on extracting data from existing data sets: NA. Data Element: Density • Need: Exposure measure. • Potential source for data element: Forward video. • Desired accuracy: NA. • Resolution: NA. • Comments on extracting data from existing data sets: The number of oncoming vehicles, of vehicles that passed by the subject vehicle, or of vehicles that the subject vehicle passes can be counted using the forward and side imagery. Density can be calculated from the number of vehicles encountered over a specific distance. Density is a good measure of roadway level of service. However, counting vehicles in the forward or side imagery is time-consuming. Data Element: Lane Departure Crash Rate • Need: Exposure measure. • Potential source for data element: State or local crash databases. • Desired accuracy: NA. • Resolution: NA. • Comments on extracting data from existing data sets: Whether or not spatially located crash databases are avail- able is the only limitation. Lane departure crashes along each vehicle trace were extracted from the Michigan Department of Transportation crash database. Crash density was calcu- lated as crashes per mile. Crash data were unavailable for Virginia, and the vehicle traces were not spatially located. Driver Factors Needed to Answer Lane Departure Research Questions Data Element: Age, Gender • Need: Needed for sampling. Included as an independent variable in statistical analyses. • Potential source for data element: Driver questionnaire at beginning of study. • Desired accuracy: NA. • Resolution: NA. • Comments on extracting data from existing data sets: Age and gender information are available in the UMTRI road- way data sets, as well as in VTTI video reduction data sets. • Limitations: NA. Data Element: Measures of Driver Riskiness • Need: Included as an independent variable in statistical analyses. • Potential source for data element: The general riskiness of each participant can be obtained from the survey at the beginning of the study. The amount of time in hard acceler- ation or the amount of time exceeding the speed limit can be calculated from vehicle data. • Desired accuracy: NA. • Resolution: One riskiness level according to the Dula Dangerous Driving Index (DDDI) questionnaire for each trip/event. • Comments and limitations on extracting data from exist- ing data sets: The subject aggression levels were collected by DDDI questionnaires at the beginning of the VTTI 100-car

38study, but the detailed results for each driver were unavail- able. Acceleration and braking data can be obtained from the data sets, but speed limit information is available neither in the VTTI nor the UMTRI databases. Data Element: Driver Distraction (e.g., Passengers, Cell Phone Usage, Eating) • Need: Included as an independent variable in statistical analyses. • Potential source for data element: Driver distractions dur- ing the trip or event can be obtained directly from the driver’s face video or from video reduction data. • Desired accuracy: Image resolution (640 × 640 pixels) for the imagery data. • Resolution: 15 Hz as the minimum and 30 Hz as preferred. • Comments on extracting data from existing data sets: Driver distractions can be extracted from UMTRI driver’s face video. The driver distraction information has already been extracted by VTTI from the driver’s face video. The eye location data that can indicate type and duration of driver distraction, as well as narrative distraction information at the time of a near crash or crash, is provided by VTTI. • Limitations: There are some missing values in the eye loca- tion data set of the VTTI data, which means eye location during some time intervals is unknown. This makes it hard to assess whether or not the driver was distracted. Data Element: Kinematic Measures Before and After Incident, Such as Acceleration, Braking, and Steering • Need: Included as an independent variable in statistical analyses. • Potential source for data element: Steering angle may be obtained from vehicle instrumentation; acceleration and braking information can be obtained from the time series data sets. • Desired accuracy: Image resolution (640 × 640 pixels) for the imagery data. • Resolution: 15 Hz as the minimum and 30 Hz as preferred. • Comments on extracting data from existing data sets: The steering, acceleration, and braking information during near- crash and crash events was extracted from VTTI video data and provided in the video reduction data set by VTTI. The acceleration and braking information can be extracted from UMTRI foot/brake pedal video. Braking information is also available in both the VTTI and UMTRI roadway data sets. • Limitations: Because there is no steering wheel sensor, information on steering is from video and is highly depen- dent on video view and quality for both VTTI and UMTRI data. Steering information for VTTI data is only available atthe time of the near-crash and crash events but not at other times when nonincident lane departure events took place. Such information is hard to obtain because the face videos from VTTI that can provide steering information are inac- cessible to us as a result of institutional review board (IRB) constraints. Data Element: Alcohol and Drug Usage • Need: Included as an independent variable in statistical analyses; however, this only refers to drug or alcohol use if present in the driver video. • Potential source for data element: Alcohol and drug use information can be obtained from video reduction data set. • Desired accuracy: Image resolution (640 × 640 pixels) for the imagery data. • Resolution: 15 Hz as the minimum and 30 Hz as preferred. • Comments on extracting data from existing data sets: Alcohol and drug usage information at the time of near- crash and crash events was extracted and provided in the VTTI video reduction data set. • Limitations: Alcohol and drug usage information for UMTRI data is inaccessible. Data Element: Driver Fatigue • Need: Included as an independent variable in statistical analyses. It should be noted that it is not simple to determine what constitutes driver fatigue and how fatigue should be measured. Some researchers have suggested that the variable that should be measured is drowsiness. This variable is only mentioned here because it has been included in other natu- ralistic driving studies. • Potential source for data element: Indicators of driver fatigue can be obtained from driver’s face video or the video reduction data set. • Desired accuracy: Image resolution (640 × 640 pixels) for the imagery data. • Resolution: 15 Hz as the minimum and 30 Hz as preferred. • Comments on extracting data from existing data sets: Indicators of driver fatigue information at the time of near- crash and crash events were extracted and provided in the VTTI video reduction data set. Eye location data from the VTTI database also provides information about the time and duration of drivers’ eye closures, which can indicate fatigue during the trip. Driver fatigue indicators can also be extracted from UMTRI driver’s face video. • Limitations: NA. Data Element: Lane Departure Intention • Need: Drivers may intentionally leave their lane for a num- ber of reasons that may result in a conflict or crash (e.g.,

39driver goes around stalled vehicle on shoulder or passes the vehicle). • Potential source for data element: Data can only be obtained from lane position tracking algorithms and asso- ciated data streams such as forward video. • Desired accuracy: NA. • Resolution: NA. Other Observations Regarding Data Elements The VTTI data sets provide gender and age information of the primary drivers but not for secondary drivers who also used the subject cars. Because there is no driver’s face video provided by VTTI, demographic information of the secon- dary drivers could hardly be obtained. This is not a problem for UMTRI data, which present all needed demographic information in both the data sets and report appendix. Summary of Vehicle, Roadway, Environmental, and Exposure Factors The factors discussed in the preceding sections are summarized in Tables 4.1 to 4.5. An indication of the priority for the data element is also provided.Table 4.1. Vehicle Factors Data Element Data Stream Accuracy Frequency Priority Vehicle position (latitude, longitude) Distance between vehicle and strikable objects Lane position, lane offset ax and vx ay and vy Pitch, roll, yaw Distance between vehicles GPS Spatial location of vehicle/objects or radar Measured by lane position tracking system using forward- or other outward-facing video, GPS, and other data streams Accelerometer or On-Board Diagnostics (OBD) Accelerometer or OBD Accelerometer Imagery, radar Best possible (±6.6 ft [2 m]) ±3.0 ft (0.914 m) ±0.1 ft (0.305 m) ±0.1 ft/s2 and 0.1 ft/s (0.0305 m/s) ±0.1 ft/s2 and 0.1 ft/s (0.0305 m/s) ±3.0 ft (0.914 m) 10 Hz NA 10 Hz 10 Hz 10 Hz 10 Hz NA High High High High High High HighReview of Planned Data Collection for Full In-Vehicle Naturalistic Driving Study The team first reviewed the various data sets that are currently available, as described in the previous sections, and com- mented on their adequacy for answering the lane departure research questions. The next step was to review any relevantinformation from SHRP 2 Safety Projects S03 (Roadway Mea- surement System Evaluation), S04A (Roadway Information Database Developer, Technical Coordination, and Quality Assurance for Mobile Data Collection), and S05 (Design of the In-Vehicle Driving Behavior and Crash Risk Study) that was available. Existing documentation that describes the instru- mentation package, roadway data collection protocol, and description of other data sources for the full in-vehicle natu- ralistic driving study were reviewed, including the following: • Design of the In-Vehicle Driving Behavior and Crash Risk Study. Task 4: Sample Design Interim Report. SHRP 2 Safety Project S05. Dingus et al. Virginia Tech Transportation Institute. November 2007. • “Gaps Identified in the SHRP 2 Safety Program, Relative to Project S04.” White paper. SHRP 2 Safety Project S03. John E. Hunt and Anita P. Vandervalk. Applied Research Associ- ates, Inc., and Cambridge Systematics, Inc. Received Febru- ary 2009. Includes appendices. • “Sampling Thoughts for S05 Following the July 2007 SHRP Safety Research Workshop.” White paper. Jim Hedlund. July 28, 2007. • SHRP 2 Safety Program Data Processing Steps. Draft. Novem- ber 4, 2008. • Design of the In-Vehicle Driving Behavior and Crash Risk Study. Task 9: Data System Interim Report (Task 6: Driver Face and Other Video Recording and Processing). SHRP 2 Safety Project S05. Dingus et al. Virginia Tech Transporta- tion Institute. September 3, 2008. • Design of the In-Vehicle Driving Behavior and Crash Risk Study. Task 9: Data System Interim Report (Task 7: Data Items and Instrumentation Package Specifications). SHRP 2 Safety

40Table 4.2. Roadway Factors Data Element Data Stream Accuracy Frequency Priority Lane width Roadway and shoulder surface type Shoulder and median width Number of lanes, access control, presence and type of median Curve length and radius Superelevation, lane cross slope Curve direction Distance between successive curves Type and characteristics of curve spirals Amount of grade (percent), length of grade, and location and characteristics of the crown and crest vertical curve Signing Number of driveway or other access points Presence and type of edge and centerline rumble strips Roadway delineation Location and type of roadside objects Mobile mapping, forward video Roadway data sets, mobile mapping, forward video Roadway data sets, mobile mapping, forward video Mobile mapping, aerial imagery Mobile mapping or aerial imagery Mobile mapping van Forward imagery, aerial imagery Mobile mapping data, aerial imagery Roadway data sets, mobile mapping, forward imagery Roadway data sets, mobile mapping Existing sign inventories, mobile mapping, forward imagery Mobile mapping, aerial imagery, forward imagery Roadway data sets, mobile mapping, forward imagery Forward imagery Mobile mapping data, aerial imagery ±0.328 ft (0.1 m) NA ±0.5 ft (0.15 m) NA ±25 ft (7.62 m) ±0.5%, 0.1% NA ±25 ft (7.62 m) ±25 ft (7.62 m) 0.5% for grade and ±25 ft (7.62 m) ±6.6 ft (2.0 m) NA NA NA ±3.0 ft (0.914 m) 10 Hz NA 10 Hz Once per mile or when characteris- tics change Once per curve Several times per mile Collected for each curve Once per curve Once per curve Begin and end points of grade change Once per sign As needed Start and end of rumble strip Once per mile or as situation changes As they occur High High High Medium High Medium High Medium Medium Medium High Medium High Medium MediumTable 4.3. Environmental Factors Data Element Data Stream Accuracy Frequency Priority Roadway surface condition Ambient condition Ambient lighting including street lighting Visibility Archived RWIS data, forward- or other outward-facing imagery, status and frequency of wiper blades, outside temperature Archived weather information, forward imagery Sun angle, dawn, dusk, day, night indicator can be obtained from time stamp data and U.S. Naval Observatory astronomical data, subjective measure from forward imagery Forward- or other outward-facing imagery Will be qualitative measure Will be qualitative measure Will be qualitative measure Will be qualitative measure 10-min intervals or if conditions change 10-min intervals or if conditions change Once per mile or as conditions change Once per mile High Low, if surface condition is collected Medium High

41Table 4.4. Exposure Factors Data Element Data Stream Accuracy Frequency Priority AADT Roadway data sets Most current year NA Medium available Time into trip Vehicle data stream NA 10 Hz Medium Amount of time driving on various Vehicle data stream NA 10 Hz High roadway types under different conditions Density Forward/side imagery NA NA High Lane departure crash data State or local crash databases NA NA MediumTable 4.5. Driver Factors Data Element Data Stream Accuracy Frequency Priority Age and gender Measures of riskiness Driver distraction Driver action before and after incident Alcohol and drug usage Driver fatigue aNot applicable because majority of measures are qualitative. Driver questionnaire Questionnaire, roadway data sets Face imagery, video reduction data sets Face imagery, roadway data sets, video reduction data Video reduction data Face imagery, video reduction data NAa NAa Image resolution (640 × 640 pixels) Image resolution (640 × 640 pixels) Image resolution (640 × 640 pixels) Image resolution (640 × 640 pixels) NA Once per trip/event 15 (minimum), 30 (preferred) Hz 15 (minimum), 30 (preferred) Hz 15 (minimum), 30 (preferred) Hz 15 (minimum), 30 (preferred) Hz High Medium High Medium Medium MediumProject S05. Dingus et al. Virginia Tech Transportation Institute. September 3, 2008. • Design of the In-Vehicle Driving Behavior and Crash Risk Study. Task 9: Data System Interim Report (Task 7: Data Items and Instrumentation Package Specifications—Appendices A and B). SHRP 2 Safety Project S05. Dingus et al. Virginia Tech Transportation Institute. September 3, 2008. The following sections describe the research team’s under- standing of relevant data collection sensors/techniques and the expected accuracy and frequency of data collection. The data elements were compared with the requirements set out in the section “Review of Roadway, Environmental, and Vehicle Data Elements Available in Existing Naturalistic Driving Study Data” (p. 28). The adequacy and limitations of the methods, accuracy, and data collection frequency for answering lane departure research questions are discussed. The following sum- marizes information for roadway, environmental, and vehicle data elements. This review is based on the information avail- able to the research team as of January 2010.Review of Planned Data Elements from Mobile Mapping for Full-Scale Naturalistic Driving Study The team reviewed a white paper developed for Safety Project S03 entitled “Gaps Identified in the SHRP 2 Safety Program, Relative to Project S04” by John E. Hunt, Applied Research Associates, Inc., and Anita P. Vandervalk, Cambridge System- atics, Inc. The white paper contains an appendix that lists items to be included in the data collection demonstration that poten- tial vendors for Safety Project S04B were asked to collect. The team reviewed the data elements along with the expected accuracy and frequency, and the following section provides its comments about how well the data would answer the lane departure research questions discussed in previous sections. The data elements for the data collection demonstration (rodeo) as indicated by Hunt and Vandervalk are provided in Tables 4.6 to 4.10. A description of the data features and data elements, along with the expected frequency and/or accuracy,

42Table 4.6. Final Rodeo Asset Data Elements Adequate for Lane Departure Research Feature Data Elements Frequency Accuracy Questions Barrier (presumably this includes median barriers and guardrail) On-street parking Pavement markings Roadside obstacles Rumble strips Sidewalk Signs Street lighting Barrier type, post type, end treatment type and location (roadside or median) Begin and end location Barrier height Presence of offset bracket or rub rail Begin, end of parking Location (right, left, both) Begin and end point, type, centerline marking type Marking offset Retroreflectivity Location of special pavement marking Description of special pavement marking Presence and location of raised pavement markers Type Offset Location Location Begin and end Offset from edge of lane Begin and end Separated from road Support type, multiple signs, and sign type Support location Location 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% NA ±3 ft (0.914 m) ±1 in. (0.025 m) ±3 ft (0.914 m) ±1 in. (0.025 m) ±0.1 m cd/m2 ±3 ft (0.914 m) ±0.25 ft (0.076 m) ±3 ft (0.914 m) ±3 ft (0.914 m) ±1 in. (0.025 m) ±3 ft (0.914 m) ±3 ft (0.914 m) ±3 ft (0.914 m) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yesis provided. Data elements not necessary to answer lane depar- ture research questions are not included. The positional accuracy for various data elements is listed as ±3 ft (0.914 m). Position of the vehicle and objects is most important in determining total transfer capability or distance to collision (DTC). If a vehicle were traveling 60 mph (80.67 ft/s) and the nearest strikable fixed object were located within ±3 ft, the error in calculating total transfer capability would be 3 ft ÷ 80.67 ft/s = 0.0372 s. For a vehicle traveling at 35 mph, the error would be 0.058 s. An error of 0.1 s would be acceptable for cal- culating time to collision, so the stated accuracy of the data col- lection is within that range. The distance error would be ±3 ft (0.914 m). However, the error from the vehicle position is not considered at this point.With the exception of lane width, all data elements met or exceeded the desired accuracy that was determined to be nec- essary to answer lane departure research questions as defined in the section “Review of Roadway, Environmental, and Vehi- cle Data Elements Available in Existing Naturalistic Driving Study Data” (p. 28). The accuracy of lane width is stated as ±0.5 ft (0.152 m). Lane width is critical in determining whether or not a vehicle that leaves the lane edge has crossed onto an unpaved shoulder; an accuracy of ±0.328 ft (0.1 m) would be preferable. The lane position tracking software that is part of the instrumentation package will measure lane width as well and use this information to determine vehicle position within its lane. The lane tracking software is expected to be less accu- rate than the mobile mapping data collection method. It will

43Table 4.7. Final Rodeo Geometric Data Elements Adequate for Lane Departure Research Feature Data Elements Frequency Accuracy Questions Grade Cross slope Curvature Direction and percent Location Location Roadway cross slope Clear zone cross slope Clear zone width Horizontal PC (point of curvature) and PT (point of tangency); vertical PC and PT Horizontal curve, vertical curve, and transition curve length Horizontal and vertical curve radius Horizontal curve super elevation Vertical curve type, presence of transition curve Stopping sight distance 100% ±0.5% ±3 ft (0.914 m) ±3 ft (0.914 m) ±0.01% ±0.25% ±0.5 ft (0.152 m) ±3 ft (0.914 m) ±2 ft (0.61 m) ±25 ft (7.62 m) ±0.05% ±10 ft (3.05 m) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesTable 4.8. Final Rodeo Intersection Data Elements Adequate for Lane Departure Research Feature Data Elements Frequency Accuracy Questions Intersection configuration and dimensions Traffic control Signal Stop control Type, number of approaches, number of through lanes, presence of channeliza- tion, number of left-turn lanes, number of right-turn lanes, presence of cross- walks, presence of illumination Location Skew Length of left- or right-turn lanes Type Type, pedestrian signal head present Location Type, presence of flashing beacon 100% 100% 100% ±3 ft (0.914 m) ±0.5° ±2 ft (0.61 m) ±3 ft (0.914 m) 100% Yes Yes Yes Yes Yes Yes Yes Yes

44Table 4.9. Final Rodeo Pavement Condition Data Elements Adequate for Lane Departure Research Feature Data Elements Frequency Accuracy Questions Pavement edge Pavement profile Skid Amount of pavement edge drop-off Location Roughness measures Critical pavement failure Macrotexture Reported at 0.1 mile intervals ±0.5 in. (0.013 m) ±3 ft (0.914 m) ±10 in./mile 100% Yes Yes Yes Yes YesTable 4.10. Final Rodeo Roadway Data Elements Adequate for Lane Departure Research Feature Data Elements Frequency Accuracy Questions Bridges Driveway Lanes Median Rail crossings Ramps Shoulder Begin and end Presence of approach slab or bridge rail Offset Location Type Number or special lane function type Lane width Location, lane add point, lane drop point Type Location Width Location Number of tracks, control type, crossing number Grade of approach or leave side Location Type of terminal, type of section Type Paved width, shoulder total width Location 100% 100% 100% 100% 100% 100% 100% ±3 ft (0.914 m) ±2 ft (0.61 m) ±3 ft (0.914 m) ±0.5 ft (0.152 m) ±3 ft (0.914 m) ±3 ft (0.914 m) ±0.5 ft (0.152 m) ±3 ft (0.914 m) ±0.5% ±3 ft (0.914 m) ±0.5 ft (0.152 m) ±3 ft (0.914 m) Yes Yes Yes Yes Yes Yes No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

45be important to compare actual lane width to what is collected by the lane position tracking system to serve as a check. Hence, accurate measurement of lane width is important. It should be noted that the upcoming project SHRP 2 Safety Project S04A will make additional decisions about what data will be collected by the mobile mapping vehicles in Safety Proj- ect S04B. This will change the scope of what is being collected. Review of Planned In-Vehicle Instrumentation Package and Available Data Elements The following sections summarize a review of the instrumen- tation package that is planned for the full-scale naturalistic in-field driving study. The sections provide a summary of information that was available as of January 2010. It should be noted that the final specifications for the data acquisition system (DAS) are not yet available, and some dif- ferences will be present between what has been reviewed here and what is available with the final DAS. The in-vehicle instrumentation package is expected to con- sist of the following sensors/elements (Dingus et al., 2008a, Task 6; Dingus et al., 2008b, Task 7): • Two forward-looking cameras; • Three rear-looking cameras; • GPS; • Incident button; • Microphone; • Speaker; • Alcohol sensor; • Light sensor; • Bluetooth radio to communicate with the forward radar; • Acceleration and orientation sensor; • On-Board Diagnostics (OBD) II; • Forward radar; and • Machine vision capabilities, including lane position and edge sensing, eyes-forward monitor, and traffic signal state. Each sensor or element of the DAS that will provide relevant information for answering lane departure research questions is discussed below. The following information is provided for each sensor/element: • Description; • List of data elements that are best collected from that sensor/ element or are only available from that source, as well as potential data elements that could be collected when they cannot be obtained from other sources; • Expected accuracy; • Expected resolution of data collection; and • Limitations in obtaining the type, amount, or quality of data necessary.Sensor/Elements Related to Roadway, Environmental, and Vehicle Factors Sensor/Element: GPS (Dingus et al., 2008b, Task 7, Appendices A and B) • Necessary data elements from sensor/element: Position (latitude, longitude, altitude), heading. • Data from sensor/element as secondary source: Forward and side speed and acceleration (best from On-Board Diagnostics [OBD]). • Accuracy: Not stated. • Resolution: 10 Hz. (It has recently been brought to the research team’s attention that GPS data may be collected at 1 Hz.) • Comments and limitations: Accuracy was not stated but is highly relevant in determining vehicle position. Accuracy is necessary to link the vehicle to the appropriate roadway and extract corresponding roadway data elements. It will also be necessary to determine vehicle position when a lane depar- ture event occurs. The best source of distance data between the vehicle and other objects will be radar, but the vehicle position from the GPS will also be necessary. Sensor/Element: Forward Video (Dingus et al., 2008a, Task 6) • Description: Two forward videos are expected for the full- scale study, which will provide a forward view of the road- way from the perspective of the driver’s field of view. One video will show a wide forward view of approximately 60° (primary forward view), and the other will be a narrow view of approximately 25° (secondary forward view). The sec- ondary forward view will be zoomed/tilted to provide sup- port for assessing traffic signal state. Both videos will be taken with color cameras. • Necessary data elements from sensor/element: The fol- lowing data elements are those where the forward camera is the best or only available source of information: – Sequence of events; – Ambient conditions (e.g., clear, raining, snowing); – Road surface condition (e.g., dry, wet, snow covered, surface irregularities); – Oncoming traffic density; – Identification of surrounding objects when vehicle engages in lane departure; – Identification of pedestrians; – Visibility; – Lane position tracking; – Verification of lane departure; – Curve direction; – Status of lane markings (e.g., highly visible, obscure); and – Signal state.

46It may be necessary to determine sign location (signs may be placed or removed after a roadway is scanned using the mobile mapping system); it will also be possible to provide a qualita- tive measure of sign condition and retroreflectivity. • Data from sensor/element as secondary source: While it may not be practical to obtain roadway data from the for- ward video, it can serve as both a check and a source when the data cannot be obtained from other sources. Data ele- ments that can be obtained include the following: – Indication of whether the subject vehicle is following another vehicle and how closely (subjective measure only); – Roadway data element type (e.g., shoulder type, presence and type of guardrail, type of signing, driveways, type of intersection control, presence and type of rumble strips); – Although not ideal, measurement of certain roadway elements is possible when the forward image is calibrated (e.g., shoulder width, distance to nearby fixed objects); and – Qualitative measure of ambient lighting. • Accuracy: It was stated that the forward view video resolu- tion would be at least 320 × 240 pixels. • Resolution: 10 Hz. Video will be stored in quad-format stor- age at one-fourth of the resolution, as shown in Figure 4.11.Source: Dingus et al., 2008a, Task 6. Figure 4.11. Combined video views from VTTI.An example of actual images in the stored format is shown in Figure 4.12.Figure 4.12. Example of video view.• Comments and limitations: Use of color cameras for the forward views will provide enhanced ability to distinguish roadway features. Color video will be particularly useful in identifying roadway surface condition and critical for deter- mining traffic signal state. It will also be useful in determin- ing condition of lane lines and signs.

47The wide-view forward video appears to have a fish-eye view that distorts objects and perspective. Although distance and identification of objects may come from other data sources, the fish-eye view could affect the ability to detect nearness of objects, identify objects, and identify sequence of events. The distortion appears worst at the edges of the image. This distor- tion could affect a data reductionist’s ability to determine run-off-road (ROR) activities. It will also be important to know if the original imagery data will be retained and will be available. It was stated that the for- ward video would be collected with a minimum resolution of 320 × 240 pixels. The images will be cropped and compressed into a view as shown in Figure 4.12. If the forward views are compressed and/or cropped, it may be advantageous for some applications to have the ability to view the forward image in its original form. Sensor/Element: Rear-View Video (Dingus et al., 2008a, Task 6; Dingus et al., 2008b, Task 7, Appendices A and B) • Description: The rear view provides an image of the road- way from the rear window of the vehicle, generally reflect- ing the driver’s rear-view perspective. The view should include the rear driving environment and may provide some ability to identify presence of back seat passengers. • Necessary data elements from sensor/element: Presence of other vehicles, ambient conditions if they cannot be identi- fied with the forward video. • Data from sensor/element as secondary source: NA. • Accuracy: Minimum resolution is 320 × 120 pixels. • Resolution: 10 Hz. • Comments and limitations: Given the image resolution pro- vided in Figure 4.12, it may be difficult to distinguish vehicles that are not closely following the subject vehicle. This infor- mation may not be relevant for lane departures, however. The data collection resolution of 10 Hz is adequate. Sensor/Element: External Right-Side View (Dingus et al., 2008a, Task 6) • Description: The right-side view provides an image of the roadway from the right of the driver, reflecting the driver’s right-side view. • Necessary data elements from sensor/element: Presence of vehicles to the subject vehicle’s right rear; presence of objects to right rear. • Data from sensor/element as secondary source: NA. • Accuracy: Minimum resolution is 320 × 120 pixels. • Resolution: 10 Hz.• Comments and limitations: The data collection resolution of 10 Hz is adequate. The image should be able to provide information on the presence of another vehicle during a right-lane departure. However, it may be difficult to distin- guish relative position given the resolution indicated in Figure 4.12. Sensor/Element: Instrument Panel View (Dingus et al., 2008a, Task 6) • Description: Provides an over-the-shoulder view of the steering column and instrument panel. • Necessary data elements from sensor/element: Relevant for driver factors. • Data from sensor/element as secondary source: May pro- vide indication of change in steering angle. • Accuracy: Minimum resolution is 320 × 240 pixels. • Resolution: 10 Hz. • Comments and limitations: Should be adequate. Sensor/Element: Machine Vision Lane Tracking (Dingus et al., 2008a, Task 6; Dingus et al., 2008b, Task 7, Appendices A and B) • Description: VTTI Road Scout (VRS) uses machine vision to determine the presence and type of lane lines and will cal- culate a vehicle’s position within its traffic lane. The system has the capability to determine when a vehicle is in the lane, crosses a solid marking, crosses a dashed line during lane changes, or when a lane change is aborted. The system is self-calibrating. It was tested on a four-lane divided road with a grass median, a four-lane divided road with left curb (no lane line), a two-lane road with pavement mark- ings, a two-lane road with no pavement markings, and a two-lane gravel road. It is expected to work in the follow- ing environments: – Interstates; – Lined highways; – Tangent and curved sections; – Under nighttime driving with or without overhead illumination; – In inclement weather when lane lines are visible; – On blacktop and concrete; and – Situations where only the centerline is present or visible. The system could not determine vehicle position on gravel or on two-lane, unmarked suburban segments. It was not clear if the system had been tested on two-lane rural segments. It can detect an upcoming curve and detect differences in lane marking type (none, double line [solid or dashed], single line [solid or dashed], road gutter, road edge, and raised pavement markings).

48System output includes the following: – Lane offset, which is the position of the center line of the vehicle with respect to the centerline of the roadway; – Lane width, determined by estimating the width of the marked lane in inches; – Line distance, which is the distance from the center of the vehicle to the right- or left-lane line; and – Probability percentages, which is a measure of the like- lihood that the pavement markings exist (serves as a key indicator for the overall reliability of the system). The system is also expected to be able to estimate hori- zontal curve radius. However, the accuracy is unknown. • Necessary data elements from sensor/element: Lane posi- tion, lane location, lane changes, type of lane lines, lane width, offset from center of lane, distance of wheel from right- or left-lane boundary, angle of departure (calcu- lated), lane departure, road departure. • Data from sensor/element as secondary source: Radius. • Accuracy: (Units reported in meters.) The accuracy of lane- position offset was stated as ±0.656 ft (0.2 m) 95% of the time when lane tracking confidence is high. The system will store distance from right tire to right-lane boundary, dis- tance of left tire to left-lane boundary, and lane width with the same accuracy. The accuracy of the horizontal curvature is expected to be ±10% of actual roadway radius when lane- tracking confidence is high. • Resolution: Minimum 10 Hz. • Comments and limitations: The expected accuracy of lane-position tracking for the proposed full-scale study is expected to be ±0.656 ft (0.2 m), which is lower than the accuracy for the UMTRI data. The average tire width is around 6.5 to 9 in. (0.165 to 0.229 m). Hence, the error is within a normal tire width. While an accuracy of 0.328 ft (0.1 m) is preferable, the proposed accuracy is expected to be sufficient. Additional collection of vehicle position at 10 Hz is adequate to establish the angle of departure and offset. The radius calculated by the UMTRI lane-tracking system appeared to be inaccurate. It will be important to verify the accuracy of the VRS system in calculating curve radius. Sensor/Element: Forward Sensor/Radar (Dingus et al., 2008b, Task 7, Appendices A and B) • Description: The system will have a forward radar capable of tracking and storing information on the five objects clos- est to the vehicle. Objects can be identified and tracked for up to 200 m in front of the vehicle within ±0.324 radians (18°) within the horizontal field of view centered on the test vehicle heading. The system can identify vehicle type (car,motorcycle, truck, pedestrian/bicyclist) and will indicate what it detects to be the lead vehicle, defined as the closest vehicle occupying the same lane. It was not clear if the sys- tem can identify roadside objects other than that they are present. • Necessary data elements from sensor/element: Distance to and location of the nearest strikable object (including other vehicle); vehicle spacing as indicator of aggressive driving. • Data from sensor/element as secondary source: The type of object will likely be determined from forward video but can be confirmed with radar. • Accuracy: Distance (in meters), accuracy will be ±1.64 ft (0.5 m); vehicle target range (in meters per second), accu- racy is ±1.64 ft/s (0.5 m/s); relationship to target object (stored in a Polar or Cartesian coordinate system), accuracy is ±0.052 radians (3°) for Polar; lateral and longitudinal off- set (in meters), accuracy ±3.28 ft (1.0 m). • Resolution: Will store at a minimum of 40 Hz for each track. • Comments and limitations: The stated accuracy of ±1.64 ft (0.5 m) is adequate to measure time and distance to collision. Sensor/Element: Automated Collision Identification and Notification (Dingus et al., 2008b, Task 7, Appendices A and B) • Description: System that will continuously monitor vehi- cle sensors to determine when a potential collision has occurred. The parameters include (1) a longitudinal accel- eration of at least 3.5 g for at least 500 m, (2) lateral accel- eration of at least 3.5 g for at least 500 m, or (3) air bag deployment status. • Necessary data elements from sensor/element: Indication of collision. • Data from sensor/element as secondary source: NA. • Accuracy: NA. • Resolution: NA. • Comments and limitations: NA. Sensor/Element: Light Sensor (Dingus et al., 2008b, Task 7, Appendices A and B) • Description: Senses amount of light. • Necessary data elements from sensor/element: Amount of daytime lighting, presence and amount of nighttime lighting. • Data from sensor/element as secondary source: NA. • Accuracy: Illumination (in lux from 3 to 80,000), accuracy is ±3%. • Resolution: 10 Hz. • Comments and limitations: Appears to be adequate for project objectives.

49Sensor/Element: Sensor to Record Internal Ambient Temperature (Dingus et al., 2008b, Task 7, Appendices A and B) • Description: Record of vehicle cabin temperature. • Necessary data elements from sensor/element: Outside temperature would be useful in making some estimates about roadway surface condition. • Data from sensor/element as secondary source: NA. • Accuracy: ±1°C. • Resolution: Sampled as changes occur of at the rate of once per 5 min. • Comments and limitations: It is not clear why internal ambient temperature is being recorded and not outside ambient temperature. Sensor/Element: Acceleration and Orientation Sensor (Dingus et al., 2008b, Task 7, Appendices A and B) • Description: Records lateral acceleration, longitudinal acceleration, vertical acceleration, and yaw rate. • Necessary data elements from sensor/element: Lateral and forward acceleration, pitch, pitch rate, yaw, yaw rate. • Data from sensor/element as secondary source: NA. • Accuracy: Lateral acceleration (±0.01 m/s2), longitudinal acceleration (±0.01 m/s2), vertical acceleration (±0.01 m/s2), and yaw rate (radians/second). • Resolution: Stored at 10 Hz during normal operation and higher during high rate of acceleration. • Comments and limitations: The stated accuracy appears adequate to answer the research questions. Sensor/Element: OBD (Dingus et al., 2008b, Task 7, Appendices A and B) • Description: OBD monitors parts of the chassis, body, and accessory devices and the diagnostic control network of the car. • Necessary data elements from sensor/element: The fol- lowing data elements will be provided and are relevant for answering the research questions: – Forward speed (m/s); – Accelerator pedal position (percent); – Brake state (on/off ); – Steering wheel position (radians of rotation); – Brake pedal force (lb/in2); – Horn status (on/off); – Gear; – Headlight status (on/off, parking); – High beam (on/off); – Wiper status (intermittent, slow, high, manually activate);– Cruise control (on/off); – Seat belt status (on/off); – Front-seat passenger status (present/not present); – Wheel speed (m/s); – Automatic braking system (ABS) activation; – Air bag deployment; – Electronic stability control (indication of when active, if present); – Traction control (indication of when active, if present); – Lane departure warning system (indication of when active, if present); – Forward collision warning system (indication of when active, if present); – Distance driven during trip (in kilometers); – Turn signal status (off, left, right, hazard); and – Driver-initiated event (indication of when driver presses event button). • Data from sensor/element as secondary source: NA. • Accuracy: Accuracy was not stated for any of the relevant items. The accuracy of the OBD system is assumed to be sufficient. • Resolution: 10 Hz. • Comments and limitations: Accuracy and resolution are expected to be sufficient. No additional necessary items from the OBD were determined. Sensor/Elements Related to Driver Factors Sensor/Element: Driver-Face Video (Dingus et al., 2008a, Task 6) • Description: The face video provides an image of the driver’s face, which can indicate driver distraction and eye location. • Necessary data elements from sensor/element: Presence of driver impairments (distraction, fatigue, emotion) over time, especially before and at the time of a lane departure. A driver’s glance direction may indicate driving-related behav- iors, preincident awareness of conflict, and postincident behavior. • Data from sensor/element as secondary source: Driver identification. • Accuracy: Minimum resolution is 640 × 640 pixels. • Resolution: 15 Hz as the minimum and 30 Hz as preferred. • Comments and limitations: The data collection resolution of 15 to 30 Hz should be adequate. The image should be able to provide information on driver fatigue, emotion, and any secondary tasks other than driving, such as talking on a cell phone or with passengers and reaching for objects in the vehicle. However, some secondary tasks, such as text messaging, might not be captured clearly, since the image might not be large enough. This kind of information needs to be provided by other videos.

50Sensor/Element: Instrument Panel View (Dingus et al., 2008a, Task 6) • Description: Provides an over-the-shoulder view of the steering column and instrument panel. • Necessary data elements from sensor/element: Hands on steering wheel; secondary tasks in which the driver engages. • Data from sensor/element as secondary source: Steering behavior at the time of incident. • Accuracy: Minimum resolution is 320 × 240 pixels. • Resolution: 15 Hz as the minimum and 30 Hz as preferred. • Comments and limitations: Should be adequate. Note: The instrument panel view has been mentioned before, but not in detail. Sensor/Element: External Passenger-Side (Right-Side) View (Dingus et al., 2008a, Task 6) • Description: This view offers information regarding the traffic around the subject vehicle and can also provide infor- mation about the passenger(s) in the front and rear seats. • Necessary data elements from sensor/element: Presence of vehicles to subject vehicle’s right rear, presence of objects to right rear. • Data from sensor/element as secondary source: Can pro- vide some information about driver distraction related to passengers. • Accuracy: Minimum resolution is 320 × 120 pixels. • Resolution: 10 Hz. • Comments and limitations: Should be adequate. Note: This view has been mentioned before, but not in respect to the driver. Sensor/Element: Passive Alcohol Sensor (Dingus et al., 2008a, Task 6) • Description: Provides information on the presence of alcohol. • Necessary data elements from sensor/element: Alcohol concentration level. • Data from sensor/element as secondary source: NA. • Accuracy: ±0.01 at 0.1 level. • Resolution: Once per hour. • Comments and limitations: The sensor is only able to detect presence of alcohol in the vehicle. The sensor may be affected by air circulation within the vehicle. Judgments about whether the driver has been drinking will have to be made using other information (e.g., the presence of alcoholand the fact that the driver is the only occupant may suggest that the driver has been drinking). The system cannot indi- cate blood alcohol level. Summary The following summarizes information about review of data elements necessary and review of what is expected to be avail- able in the full-scale study. General Comments about Extracting Data from Existing Data Sources In general, most of the roadway, environmental, and vehicle data elements desired could be extracted from the UMTRI nat- uralistic driving study data and related aerial imagery, crash databases, and roadway databases. The naturalistic driving data indicated when the vehicle was traveling on a curve. When vehicle traces were overlaid with aerial images and compared, the identified curve locations were quite accurate. Two data items that were not accurate were shoulder width and curve radius. This is based on a review of the UMTRI data, as described in Appendix A. The forward imagery was adequate for all the applications for which it was used. One advantage of the UMTRI imagery over the proposed forward imagery for the full-scale data col- lection is the width of the forward view that was available. It was easy to see a large portion of the forward roadway (includ- ing all of the shoulders), identify objects to the edge of the roadway, identify ambient conditions, and confirm that a vehicle was departing its lane from the forward view. While the forward view with the proposed full-scale in-vehicle instrumentation is in color, which offers additional advan- tages, the forward view does not offer the same wide view. As a result, it may be difficult to identify roadside features. The dis- tortion of the image (in fish-eye view) in the proposed instru- mentation package is also particularly problematic. The lateral portion of the proposed imagery does, however, offer a better view of overhead features. The two forward views are compared in Figure 4.13.The lane position tracking was very useful in the UMTRI data. It was relatively simple to determine when a vehicle had left the roadway. It was also relatively easy to tell when the lane position data were “bad.” The ability to determine a vehicle’s position within its lane is critical for identifying lane departure events and answering the lane departure–related research questions. Sufficient data were not provided with the VTTI data set to make the same determinations as for the UMTRI data. In general, the image resolution was too low that it was diffi-

51(a) (b) Source: (a) UMTRI RDCW data set; (b) Dingus et al., 2008a, Task 6. Figure 4.13. Comparison of forward imagery: (a) forward view from UMTRI data set; and (b) potential forward view for full-scale data collection.cult to make out many features in the forward video. It was even difficult to tell from the forward view that the vehicle was leaving the roadway or which object was in the vehicle’s path. It is not known if the image data were reduced from their original format. The database provided with the vehi- cle trace data was adequate to extract information such as lateral acceleration. Summary Comments and Concerns about Proposed Full-Scale Data Collection Methods The following is a summary of comments or concerns that arose during the review of the instrumentation packages that will be available for the full-scale data collection effort (SHRP 2 Safety Projects S03 and S05). • Mobile mapping vans (SHRP 2 Safety Project S03) – The accuracy and resolution of data collection for all data elements (except for lane width) met or exceeded the level of accuracy that was determined to be necessary to answer lane departure research questions. – The proposed accuracy of lane width to be collected with mobile mapping vans is ±0.5 ft (0.152 m). Lane width will be calculated with the vehicle instrumentation package lane position tracking system in order to determine vehi- cle position. The lane tracking system is expected to be less accurate than the mobile mapping vans, so it will be important to have an accurate measurement from themobile mapping system as a check. A more accurate measurement of lane width would be recommended, if possible. An accuracy of ±0.25 ft (0.076 m) would be preferable. – Recent information from various SHRP 2 safety meetings during the summer of 2009 have indicated that the final mobile mapping data collection may be somewhat differ- ent from what was reviewed in this document. • In-vehicle instrumentation (SHRP 2 Safety Project S05) – Accuracy of the differential GPS was not stated but is highly relevant for determining vehicle position relative to roadway features. – The use of color imagery for the forward video is a wel- come addition and will allow objects to be distinguished more easily. Use of color cameras for the forward views will provide an enhanced ability to distinguish roadway features. It will be particularly useful for identifying road- way surface conditions and critical for determining traf- fic signal state. It will also be useful for determining the condition of lane lines, and signs and will allow identifi- cation of traffic signal state. While not as relevant to lane departures, this ability will be critical for answering inter- section questions. – The image resolution of 320 × 240 pixels for the forward view appeared adequate. – The fish-eye view for the forward view distorts objects and perspective. This is highly problematic for identify- ing objects and gauging distances. Although distance and identification of objects may come from other data sources, the fish-eye view could affect the ability to detect the nearness of objects, identify objects, and identify the sequence of events. The distortion appears worst at the edges of the image. This distortion could affect the abil- ity to measure distances to roadside objects. The distor- tion could also affect the ability to identify objects. – It was indicated that all raw data (both video and sensors) would be continuously recorded and preserved (Dingus et al., 2008b, Task 7). It may be useful to view the raw forward, back, or side video because some information may be obtained that cannot be obtained with the com- pressed-resolution images. It is important to clarify whether the data will be stored in a format that is linked to the other data and whether the data can be accessed. – The lane position tracking system is critical for addressing lane departure questions. Several experts were questioned about the level of risk of different lane departure events. They unanimously agreed that even one tire leaving the paved roadway surface onto a grass, gravel, or mixed-surface shoulder constitutes a highly dangerous situation. As a result, the lane posi- tion tracking system should be accurate and reliable

52enough to determine when one or more tires have departed the roadway surface. The planned lane posi- tion tracking system for the full-scale study has a stated accuracy of ±0.656 ft (0.2 m). While this is still within the range of a normal tire width and is likely to be ade- quate, any improvements to the planned accuracy would be beneficial. A lower level of accuracy may be accept- able where the surface beyond the lane marker is hard and level. – It is important that the lane tracking system be verified both in terms of accuracy and in situations where it may not perform well. – It was stated that a combination of in-vehicle sensors could be used to determine curve radius. It is important that the accuracy be verified. The radius measurementsreceived with the UMTRI data did not appear to be accurate. – It was not stated whether the forward radar has the abil- ity to identify roadside objects. – It is unclear why internal vehicle cabin temperature is measured and recorded but not the outside ambient temperature. – The method to determine the accuracy of each stated sensor/data collection element should be stated. For example, if radius is determined using a combination of in-vehicle sensors, the test method and results to deter- mine accuracy should be stated. – Other discussions about the in-vehicle data collection system have suggested that a head-pose tracker may be available.