Analysis of Naturalistic Driving Study Data: Roadway Departures on Rural Two-Lane Curves (2014)

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.

61 A p p e n d i x A The methodology used to reduce various roadway data fea- tures is described in the sections below. Kinematic Vehicle Factors 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 departure angle from the roadway and 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 associated data streams such as forward video. Accuracy: Not yet available from Virginia Tech Transporta- tion Institute (VTTI). Resolution: 10 Hz. Comments: The NDS DAS reports information that can be used to establish lane position. Lane-tracking units were reported as centimeters in the data dictionary, but a review of the first data set indicated this was erroneous. In a follow-up conversation with VTTI, it was determined that the units ini- tially reported are millimeters. The following variables are used to calculate lane position (see also Figure A.1): • Lane Position Offset (vtti.lane_distance_off_center): Dis- tance to the left or right of the center of the lane based on machine vision. • Lane Width (vtti.lane_width): Distance between the inside edge of the innermost lane marking to the left and right of the vehicle. Note that lane width is calculated for each 0.1-second interval and varies somewhat. • Lane Marking, Distance, Left (vtti.left_line_right_ distance): Distance from vehicle centerline to inside of left- side lane marker based on vehicle-based machine vision. • Distance from vehicle centerline to inside of left-side lane marker based on vehicle-based machine vision. • Lane Marking, Distance, Right (vtti.right_line_left_ distance): Distance from vehicle centerline to inside of right- side lane marker based on vehicle-based machine vision. • Lane Marking, Probability, Right (vtti.right_marker_ probability): Probability that vehicle-based machine-vision lane-marking evaluation is providing correct data for the right-side lane markings. Higher values indicate greater probability. • Lane Marking, Probability, Left (vtti.left_marker_ probability): Probability that vehicle-based machine-vision lane-marking evaluation is providing correct data for the left-side lane markings. Offset from lane center and distance from the right lane (RD) or left lane (LD) line are the metrics currently being used as crash surrogates. RD and LD are calculated as shown in Equations A.1 and A.2 (in meters). 2 (A.1)CLL L TD w= − − 2 (A.2)CLR R TD w= − where LD = distance from left edge of vehicle to left edge of lane line; if negative, means left edge of car is to the left of the left edge line. RD = distance from right edge of vehicle to right edge of lane line; if negative, means right edge of car is to the right of the right edge line. Tw = vehicle track width. Data element: Presence and distance between subject vehicle and other vehicles Need: Establish outcome from lane departure; used as a mea- sure of level of service. Presence of other vehicles (opposing, Methodology for Reducing Roadway Data

62 vehicles passed) can be used to determine roadway density as an exposure method. Source: Forward video. Accuracy: ±3 ft (0.914 m). Resolution: Collected as vehicle was approaching the curve. Comments: A subjective measure of distance will be obtained from the forward video, as shown in Figure A.2, but distance cannot be determined. When a conflict occurs, distance to a forward or side vehicle will be determined from the forward or side radar. However, only vehicles within the radar range can be detected. Coding: Following 0: No forward vehicle present 1: Forward vehicle present but not following 2: Following closely (less than 3 seconds apart) Roadway Factors Data element: Lane width Need: Independent variable in the statistical analysis; also needed to establish vehicle position within its lane. Source: Mobile mapping when available; lane-tracking sys- tem (varies significantly over 0.1-second intervals—could use average). Accuracy: Need to determine from mobile mapping and lane tracking. Resolution: At curve approach, PC, apex, PT. Comments: Lane width is measured by the DAS lane-tracking system and will be used when position within the lane is needed. Coding: LaneWidth, reported in meters. Data element: Shoulder width Need: Independent variable in statistical analyses. Shoulder and median width also affect potential outcomes for lane departures. Source: Mobile mapping data; may be available from road- way databases. Accuracy: ±0.5 ft (0.152 m). Resolution: At curve approach, PC, apex, PT (should be checked at several points but can be reported once). Comments: Could not be accurately measured from aerial images and is therefore not included in initial analysis, as mobile mapping data are not available. Coding: Paved shoulder width 1: Less than 1 ft 2: 1 ft to less than 2 ft 3: 2 ft to less than 4 ft 4: Greater than or equal to 4 ft Data element: Curve length and radius Need: Independent variable in statistical analyses; may also be used to assess roll hazard. Source: Mobile mapping, aerial imagery. Accuracy: ±25 ft (7.62 m) for curve length and ±10% for radius. Resolution: Once per curve. Comments: Extracted for each direction and then averaged to find one value for each curve. Figure A.1. Description of variables to calculate lane position. Figure A.2. Subjective measure of vehicle following. Source: University of Michigan Transportation Research Institute (UMTRI) Road Departure Crash Warning (RDCW) data set. Subject vehicle is closely following forward vehicle. Subject vehicle is not considered to be following forward vehicle.

63 Coding: Length of curve from PC to PT, reported in meters (Length). Radius of curve, in meters (Radius). Data element: Curve superelevation 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. Accuracy: Maximum superelevation for areas with no ice and snow is 12%; for areas with snow and ice the maximum 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: Once per curve as reported by the mobile mapping. Comments: SHRP 2 Project S04A data had both negative and positive values. Coding: Extracted once per curve for each lane. Superelevation, in percent (Super). Data element: Driving direction Need: Independent variable in statistical analyses; also impor- tant for determining the potential outcome of a noncrash lane departure. Source: Aerial imagery and forward view. Accuracy: NA. Resolution: Should be indicated once per curve. Comments: None. Coding: Direction of travel (Cardinal) 0: N/S 1: E/W 2: NE/SW 3: NW/SE Direction of curve from perspective of driver (Direction) 0: Outside/left-hand 1: Inside/right-hand Data element: Distance to upstream curve, distance to downstream curve from perspective of driver (meters) Need: Drivers may negotiate curves differently if they have traveled for some distance between curves instead of having negotiated a series of curves. Also used as an independent variable in statistical analyses. Source: Aerial imagery. Accuracy: ±25 ft (7.62 m). Resolution: Upstream and downstream per curve. Comments: None. Coding: Distance to upstream curve from perspective of driver, in meters (DistUP). Distance to downstream curve from perspective of driver, in meters (DistDown). Curve type 0: Individual curve 1: S-curve (less than 600 ft between subsequent curves) 2: Compound curve (0 ft between 2; the PT and PC of sub- sequent curves in the same direction) Data element: Speed limit, curve advisory, chevrons, and W1-6 signs Need: Independent variable in statistical analyses. Source: • Speed limit and curve advisory speed limit from mobile mapping. • Forward video/Google/forward view mobile mapping for remaining. 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 that were present on a curve, but it is not necessary to know exactly where each sign is located. It is also assumed that all signs are compliant with National Cooperative Highway Research Pro- gram (NCHRP) 350 so that they would not need to be consid- ered as strikable fixed objects when determining 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: None. Coding: Tangent speed limit (SpdLimit), in miles per hour. Advisory speed (Advisory), in miles per hour, or 999 if no advisory speed limit exists. Presence of chevrons (Chevrons) 0: Not present 1: Present Presence of curve advisory sign 0: Not present 1: Present Presence of W1-6 sign 0: Not present 1: Present

64 Data element: Number of driveways or other access points Need: Traffic entering and exiting the traffic stream can affect vehicle operation. This traffic would be included as an inde- pendent variable in statistical analyses. Source: Aerial imagery and forward imagery. Accuracy: NA. Resolution: Number in the upstream, curve, and downstream. Comments: Four-way intersections counted as one cross street. Coding: Number of driveways at approach, within curve, at exit. Cross streets (CrossStreets), in points per section through length of curve and tangents. Driveways (Dwys), in driveways per section through length of curve and tangents. Data element: Presence of edge or centerline rumble strips Need: Independent variable in statistical analyses; also needed to establish outcome of lane departure. Source: Forward video and Google Street View. Accuracy: NA. Resolution: Curve approach and in curve. Comments on extracting data from existing data sets: Only presence of rumble strip could be extracted, not distance from road. Coding: Type of rumble strip (RS) 0: No rumble strip present 1: Edge line rumble strips only (see Figure A.3) 2: Centerline rumble strips only 3: Centerline and edge line rumble strips 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. Source: Forward view. Desired accuracy: Data is a quantitative estimate of visibility of markings. Resolution: Once per mile or as situation changes. Comments: This element needs to be current to driving situ- ation 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 due to image resolution. Coding: Presence of raised pavement markings (RPMs) 0: Not present 1: Present Roadway delineation (Delineation) 0: Highly visible 1: Visible 2: Obscured 3: Not present Figure A.4 shows an example of a subjective measure. Data element: Roadway furniture Need: Necessary to determine how roadside makeup affects driving; also how roadway furniture may affect the severity of a lane-departure crash. Source: Forward view. Accuracy: NA. Resolution: Once per curve just upstream of PC looking at curve ahead for roadway furniture rating; once per curve at any location for presence of guardrail. Coding: Presence of guardrail 0: Not present 1: Present Roadway furniture 1: Little to no roadway furniture 2: Moderate roadway furniture 3: Large amount of roadway furniture Figure A.5 shows an example of a subjective measure. Data element: Sight distance Need: The distance at which the curve is first visible will have an effect on where a driver reacts to the curve and could play a role in lane departures. Source: Forward view and time series data. Accuracy: NA. Resolution: Once per direction per curve. Source: DAS forward imagery. Figure A.3. Presence of edge-line-only rumble strips.

65 Comments: This was calculated once per curve using the best forward video available. At times, night was the only condition to assess sight distance of the curve. Timestamp at which curve could first be seen was recorded and then used to find corre- sponding distance upstream in time series data. Coding: Distance in meters to PC. environmental Factors This section summarizes environmental factors necessary to address lane-departure research questions, indicates poten- tial 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 (presence of roadway irregularities such as potholes) Need: Independent variable in statistical analyses; may also affect potential outcome of lane departure. Source: Forward or other outward facing video, status and frequency of wiper blades, outside temperature if available, roadway weather information system (RWIS) data if archived. Accuracy: Measure is subjective and therefore not applicable. Resolution: At curve approach, in curve. Comments: None. Coding: Roadway surface condition (PaveCnd) 0: Normal surface condition, no obvious damage present 1: Moderate damage 2: Severe damage, presence of potholes Figure A.6 shows an example of a subjective measure. Data element: Environmental conditions such as raining, snowing, cloudy, clear (may not correspond to roadway surface condition) Need: Independent variable in statistical analyses; may affect sight distance and is related to visibility. Source: Forward imagery or archived weather information, ambient temperature probe. Accuracy: Subjective measure. Resolution: Once per vehicle trace. Comments: A general assessment of environmental condi- tions can be obtained from the forward video (Figure A.7). Even with wiper position, it is difficult to tell how heavy Figure A.4. Subjective measure of lane marking condition using forward imagery. Pavement markings indicated as “highly visible.” Pavement markings indicated as “visible.” Right pavement markings indicated as “obscured.” Source: Forward video and UMTRI RDCW data set.

66 Source: DAS forward imagery. Little to no roadway furniture. Moderate roadway furniture. Large amount of roadway furniture. Figure A.5. Subjective measure of roadway furniture. rainfall is. Archived weather information can provide general information for an area but cannot tell the exact environ- mental conditions for the location where the subject vehicle is located. Coding: Roadway surface condition (Surface) 0: Dry pavement surface 1: Pavement wet but not currently raining 2: Wet and light rain 3: Wet and heavy rain 4: Snow present but road is bare 5: Snow along road edge and/or centerline 6: Light snow on roadway surface 7: Roadway surface covered Data element: Ambient lighting Need: Independent variable in statistical analyses. Source: Derived from sun angle, twilight, and forward view. Accuracy: Subjective measures. Resolution: Once per trace or as conditions change. Comments: A relative estimate of ambient lighting can be obtained in most cases from the forward imagery. The limita- tions are that it was difficult during high cloud cover or low visibility to subjectively estimate ambient lighting. Coding: Ambient lighting (Lighting) time of day and lighting 0: Daytime 1: Dawn/dusk

67 Figure A.6. Subjective measure of roadway pavement surface condition using forward imagery. Pavement condition indicated as “normal.” Pavement condition indicated as “moderate.” Source: DAS forward imagery. Figure A.7. Pavement surface condition from forward imagery. Pavement surface condition: snow present but roadway bare. Pavement surface condition: wet but amount of water cannot be determined. Surface irregularities. Source: UMTRI RDCW data set. 2: Nighttime, no lighting 3: Nighttime, lighting present Data element: Visibility Need: Independent variable in statistical analyses; serves as a measure of sight distance and can also indicate surface conditions. Source: Forward view is the only reasonable data source. Accuracy: Subjective variable. Resolution: Once per trace. Comments: This element is available from forward imagery. In some cases it may be difficult to tell whether visibility or image resolution causes securement, as shown in Figure A.8. The source of decreased visibility could not be determined. Low visibility is shown in Figure A.9, but it is unknown if the source is fog, smoke, or dust. Coding: Visibility 0: Clear 1: Reduced visibility 2: Low visibility

68 Source: DAS forward imagery. Figure A.8. Reduced visibility may be due to sun angle or image resolution. Figure A.9. Low visibility appears to be due to fog. Source: DAS forward imagery. exposure Factors This section summarizes exposure factors necessary 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: Density Need: Exposure measure. Source: Forward video. Accuracy: NA. Resolution: Number of vehicles on approach, within curve, at exit. Comments: The number of oncoming vehicles, vehicles passed by the subject vehicle, or vehicles that the subject vehicle passes can be counted using the forward and side imagery. Density can be calculated knowing the num- ber of vehicles encountered over a specific distance. Den- sity is a good measure of roadway level of service. However, counting vehicles in the forward or side imagery is time- consuming. Coding: Number of vehicles passing subject vehicle during period (Density), in vehicles per meter, calculated through curve.