Sag Vertical Curve Design Criteria for Headlight Sight Distance (2013)

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.

56 CHAPTER 6 VISIBILITY EXPERIMENTS Two human-subjects experiments were conducted to determine the effects of headlamps and sag vertical curves on visibility. The first experiment, called the Smart Road study, examined the effects of varying types of headlamps. The second experiment, called the Public Road study, examined the effects of varying sag vertical curves. The results of these experiments were then used to determine the practicality of the proposed policy changes suggested in Chapter 5. The methods, results, and discussion of the Smart Road study are described below, followed by that of the Public Road study. SMART ROAD STUDY The purpose of the Smart Road study was to determine if modern headlamp designs have diminished visibility in sag vertical curves as compared to the more traditional headlamps likely used in the development of AASHTO’s guidelines. Smart Road Experimental Design This experiment took place on the Virginia Smart Road, and compared the performance of several different types of headlamps in sag vertical curves. The study used a 6 (Headlamp) x 3 (Object) x 4 (Sag Vertical Curve) mixed-factors design. The six headlamps used either a halogen or HID light source, and used either a standard, VOL, or VOR beam pattern. One headlamp used a high beam setting. The objects which participants identified were either a pedestrian dressed in denim clothing, a 7-inch-square piece of wood painted gray (called a target), or a speed limit sign. Participants identified objects on four types of sag vertical curves: flat (i.e., no curve), large curve, small curve, and a sag vertical curve which also had a horizontal curve to the left termed the “left curve.” Table 22 shows the design matrix. Cells marked with an “X” represent scenarios which were tested. Every participant observed every test scenario. Table 22. Smart Road Study Experimental Design Matrix Pedestrian Target Speed Limit Sign Headlamp Flat Large Curve Small Curve Left Curve Flat Large Curve Small Curve Left Curve Flat Large Curve Small Curve Left Curve HHB X X X X X X X HLB X X X X X X X VOLHID1 X X X X X X X VOLHID2 X X X X X X X VORHAL X X X X X X X VORHID X X X X X X X HHB - halogen high beam; HLB - halogen low beam; VOLHID – Visually Optically Aligned Left HID; VORHAL - Visually Optically Aligned Right Halogen; VORHID - Visually Optically Aligned Right HID

57 Independent Variables Several independent variables were manipulated or controlled for this experiment. They are listed below. Between-Subjects Variables • Gender (2 levels): Female, Male. The gender independent variable was chosen in order to generalize the results of this study to a broad user population. This factor was used for balance only; it was not used in the data analysis. • Age (2 levels): Younger (21-34), Older (65+). The younger and older age groups were selected to investigate the changes in vision and perception that may occur with increasing age. Within-Subjects Variables • Headlamp (6 levels): Halogen High Beam (HHB), Halogen Low Beam (HLB), two Visually Optically Aligned Left HIDs (VOLHID1 and VOLHID2), Visually Optically Aligned Right Halogen (VORHAL), and Visually Optically Aligned Right HID (VORHID). The six different headlamps were selected to represent an array of the most common types of headlamp light sources and beam patterns. • Object (3 levels): Pedestrian, Target, and Speed Limit Sign. The pedestrian and target objects were selected to represent objects a motorist may encounter on or near the roadway that may require action to avoid. The speed limit sign object was used to determine if the amount of uplight from the different headlamps had an effect on a driver’s ability to read signs. • Sag Vertical Curve (4 levels): None (i.e., Flat), Large Curve, Short Curve, and Left Curve. The three different sag vertical curves selected were chosen based on what curves were available on the Smart Road test track. The non-curved, or flat, area was selected as a point of comparison for the sag curves. Dependent Variables Detection Distance The distance at which a participant could identify an object was recorded as a measure of the visibility of the object. Participants were instructed to verbally identify objects as they drove. Participants would say “pedestrian” or “target” aloud depending on the object being presented. At that moment, an in-vehicle experimenter would flag the data with a button press. When the vehicle reached the object, the in-vehicle experimenter would again flag the data with a button press. Later analysis determined the distance traveled between these two points, and this was termed the “detection distance.” Figure 21 illustrates the two points at which the data was flagged. Detection distance as defined in this report may be more closely related to recognition distance in that participants had to identify what an object was, rather than simply detect that something was there. However, as the only two objects used in this study were quite different in size, shape, and color, it is likely that recognition in this study was a relatively quick process.

58 Figure 21. Determining detection distance. The retroreflective nature of the speed limit signs made them visible from great distances. Therefore, detection distance for the signs was measured differently. Rather than identify when they could see the signs, participants were instead instructed to read aloud the number on the sign as soon as they could read it. Participants would say aloud “35” or “55” depending on the sign being presented. The distance between this point and the point at which the vehicle reached the sign determined the detection distance for the sign. Participants Twenty-five participants were selected to take part in this study. Participants were selected from two age categories: younger (18-34 years old) and older (65+). Six younger males, six older males, six younger females, and six older females participated. Recruitment occurred through the Virginia Tech Transportation Institute (VTTI) participant database as well as word- of-mouth. A general description of the study was provided to the subjects over the phone before they decided if they were willing to participate. If they were interested, subjects were then screened with a verbal questionnaire to determine whether they were licensed drivers and whether they had any health concerns that should exclude them from participating in the study. If subjects were determined to be eligible for the study, they were then scheduled to come to VTTI for participation. When subjects arrived at VTTI, they read and signed an informed consent form. Subjects were paid $20/hr and were allowed to withdraw at any point in time, with compensation adjusted accordingly. Participants’ Age and Visual Capabilities The ages of the younger participants ranged from 21 to 26 years old, with a median age of 22 years. The older participants ranged in age from 66 to 70 years old, with a median age of 67 years. All participants passed a color-blindness test, with only two participants giving a single incorrect response each. Table 23 presents the distribution of visual acuity scores. A

59 minimum of 20/40 was required for participation. Participants were allowed to wear corrective lenses if they indicated that they normally wear them while driving. Table 23. Visual Acuity Scores Visual Acuity (Snellen) Number of Participants 20/15 4 20/20 8 20/25 4 20/30 7 20/40 2 Participants’ visual acuity was also measured in the presence of glare using a Brightness Acuity Tester (BAT). The distribution of visual acuity scores for each eye and level of glare are presented in Table 24. Table 24. Visual Acuity in the Presence of Glare No Glare Low Glare Med Glare High Glare Acuity Score Left Eye Right Eye Left Eye Right Eye Left Eye Right Eye Left Eye Right Eye 20/13 1 1 1 1 1 1 1 1 20/15 0 0 0 0 2 0 1 0 20/20 3 4 6 4 4 4 0 2 20/25 6 6 5 5 4 4 6 3 20/30 6 4 6 6 2 5 5 3 20/40 6 7 1 5 5 5 4 7 20/50 3 0 5 1 5 4 4 3 20/70 and worse 0 3 1 3 2 2 4 6 Facilities and Equipment Smart Road Test Track This experiment was conducted on the Virginia Smart Road; a 2.2 mile-long, restricted- access test facility. The Smart Road provided three different sag vertical curve geometries. The first is on the primary roadway where the Smart Road Bridge transitions to the flatter turnaround area (Figure 22). The second curve is on an access road which intersects the main roadway (Figure 23). Here, a shallower curve is evident (the pictured gate was open during the experiment). A third curve was located in the upper turnaround (Figure 22). This section of roadway also had a horizontal curve.

60 Figure 22. Primary sag vertical curve on the Smart Road. Figure 23. Secondary sag vertical curve on the Smart Road. The measures of the large curve were found in the design documents for the Smart Road. The measurements of the short and left curves were made using Global Positioning System (GPS) data recorded using a GPS-enabled vehicle. The data were processed in ArcGIS™ and ArcMap™ 10 software to generate the elevation, showing the profile of the curves. These profiles were printed, and the tangent lines were found visually using a straight edge. The slopes

61 of the tangent lines were calculated to determine the algebraic change in grade while the horizontal distance between the two tangent points was used to determine the length of the curves. Table 25 summarizes this information. It also shows the minimum K value based on design speed, and the actual K value – called K reality or KR - based on the grade and length measurements, or design documents. Table 25. Sag Vertical Curve Measurements Curve Design Speed (mph) Minimum K Algebraic Change in Grade, A (%) Length, L (ft) KR (L/A) Short Curve* 25 26 4.67 200 43 Left Curve* 35 49 5.53 215 39 Large Curve 62 148 9.41 2297 244 *The design speeds were unknown for these curves, and were estimated by the research team. Visibility Objects Participants were asked to detect pedestrians and small targets while performing the study. Pedestrians were on-road experimenters dressed in denim-colored surgical scrubs (shown in Figure 24. The targets were 7” square pieces of wood with a small tab on one side, painted in 18% reflectance gray paint as shown in Figure 25. Pedestrians and targets appeared on the shoulder of the roadway, closest to the participant’s travel lane, approximately two feet from the lane’s edge-line. Pedestrians faced into the roadway to the roadway, and stepped backward away from the road when the participant vehicle came within a close proximity. Targets stood upright in a small wooden base which was also painted gray, and the face of the target was pointed parallel with the roadway toward the approaching vehicle. Figure 24. Example of Pedestrians used in the Visibility Experiment.

62 Figure 25. Example of Targets used in the Visibility Experiment. Headlamps and Test Vehicles Participants drove either a 1999 or 2000 model Ford Explorer equipped with a special headlamp mounting system. The mounting system allowed an experimenter to quickly change the headlamps during the study, so that participants could perform the study with each set of headlamps without having to switch vehicles. The headlamps used in the study are summarized in Table 26 below. The vehicle from which the headlamps originated is listed in the table, however all headlamps were placed only on the two Ford Explorers for this experiment. These were selected as being representative of the most common types of headlamps, which include headlamps with halogen or HID light sources, and with standard, VOL, or VOR beam patterns.

63 Table 26. Headlamps Headlamp Abbreviation Description Original Vehicle Photo HHB Standard Halogen High Beam 2000 Ford Explorer HLB Standard Halogen Low Beam 2000 Ford Explorer VOLHID1 Visually Optically Aligned Left HID 2001 Mercedes S500 VOLHID2 Visually Optically Aligned Left HID 2003 Mercedes E320 VORHID Visually Optically Aligned Right HID 2003 Lincoln Navigator VORHAL Visually Optically Aligned Right Halogen 2003 Lincoln Navigator The test vehicles were also equipped with data acquisition systems (DAS). The DAS recorded video both inside and outside of the vehicle and button inputs from the in-vehicle

64 experimenter, as well as network data such as speed and distance. Detection distance was determined by analysis of this data. Smart Road Experimental Method Participants were initially contacted and screened on the phone using an internal VTTI database of persons who had expressed interest in participating in research studies. When participants arrived at VTTI for participation, they first read and signed the informed consent. Participants then filled out a W9 tax form and a health history questionnaire. These were followed by vision tests for acuity, contrast sensitivity, and color blindness. Participants were only excluded from participation if visual acuity was less than 20/40 (the legal minimum to hold a Virginia driver’s license), or if they had taken any substance which might impair their ability to drive. Participants were scheduled in pairs, with each driving a different test vehicle. Once the paperwork and vision tests were complete, the participants were escorted to their assigned vehicle by an in-vehicle experimenter. Participants were then instructed to drive to the Smart Road. Once on the road, participants drove a practice lap in order to familiarize themselves with the road, the points at which they would be turning around, and where they would be stopping to have the headlamps on the vehicle changed. No detection tasks occurred during the practice lap. Figure 26 shows the layout of the Smart Road and the object and curve locations. Objects were only present when the participant vehicle was in the nearest travel lane (i.e., when the object was on the driver’s right hand side). Figure 26. Object and curve locations.

65 Participants drove 2 laps with each headlamp for a total of 12 laps. Starting from the headlamp changing area, participants drove through the flat area and across the large curve. They then turned around and traveled through those areas again in the opposite direction. Participants then turned right at the intersection onto the access road. They drove across the short curve and into a gravel parking lot. Participants then turned around, and drove back onto the Smart Road, crossing the short curve again. Participants then turned right onto the Smart Road and traveled through the upper loop where the left curve was located. Next, participants traveled to the end of the road, crossing the flat area and large curve again. Finally, after crossing the large curve and flat area traveling back up the road, participants stopped in the headlamp changing location to have the headlamps changed. This pattern was repeated for each of the six headlamps. Pedestrians in the short and left curves were seen only once for each headlamp. The timing of the two participant vehicles was such that the second participant was beginning a lap as the first participant drove past the headlamp changing area on his/her way up toward the short curve. The first participant would then wait at the end of the upper loop until the second vehicle was turning onto the access road. Once the second participant turned, the first would proceed with the next lap. This way, the vehicles never passed each other during data collection. As participants drove, they would verbally identify the objects (pedestrians or targets) as they encountered them, and read aloud the number on the speed limit sign as soon as they could read it. The in-vehicle experimenters flagged these moments in the data with a button press. They also flagged the point at which the vehicle reached the object or sign that was detected. Once both participants had completed all 12 laps, they were instructed to exit the Smart Road and return to VTTI. There, they were paid for their participation and given a receipt and a copy of the informed consent. Participants were compensated at a rate of $20 per hour. Presentation Orders Two factors limited the ability to fully balance the presentation of the headlamps. The first was the fact that the two vehicles which were being used simultaneously by participants had to share the headlamps. Thus, the headlamps assigned to one vehicle were dependent upon which headlamps were already assigned to the other vehicle. The second factor was that two of the headlamp levels used the same physical headlamps: high beam and low beam halogens. Thus, for efficiency, those two headlamp levels were always paired, though the order in which they appeared was balanced. Because of these factors, four semi-balanced orders were used (Table 27). Each participant performed the experiment with every headlamp; however, not every headlamp was used in every position (e.g., VOR HID did not appear in the second or fourth positions).

66 Table 27. Headlamp Presentation Orders Order 1 Order 2 Order 3 Order 4 Halogen Low Beam VOR Halogen VOR HID VOL HID 1 Halogen High Beam VOL HID 2 Halogen High Beam VOR Halogen VOL HID 1 Halogen Low Beam Halogen Low Beam VOR HID VOL HID 2 Halogen High Beam VOR Halogen Halogen High Beam VOR HID VOL HID 1 VOL HID 2 Halogen Low Beam VOR Halogen VOR HID VOL HID 1 VOL HID 2 Not all objects were presented in every curve. Signs only appeared on the flat section of roadway. Pedestrians and small targets were presented and counterbalanced on the flat section and in the main sag vertical curve on the Smart Road, termed the Large Curve. Because participants only observed objects in the Short and Left Curves once for each headlamp, only pedestrians were presented so that comparisons could be made between headlamps. Table 28 shows how many times each object was presented in each curve for each participant. Table 28. Number of Object Presentations by Curve per Participant Pedestrian Target 35mph Sign 55mph Sign Flat 12 12 6 6 Large Curve 12 12 0 0 Short Curve 6 0 0 0 Left Curve 6 0 0 0 Smart Road Data Analysis To investigate the importance of different aspects of the headlamps, three analyses of covariance (ANCOVA) with a significance level of 95% (α = 0.05) were used. The first analysis treated each of the six headlamps as its own factor. This was termed the headlamp analysis. The second analysis investigated the difference between halogen and HID light sources. This was termed the source analysis. Finally, the third analysis investigated the differences among different beam patterns (standard, VOL, and VOR). This was termed the pattern analysis. Table 29 shows how the headlamps were grouped by light source or beam pattern. The halogen high beams were excluded from the source and pattern analyses. Table 29. Headlamp Grouping by Light Source and Beam Pattern Beam Pattern Light Source Standard VOL VOR Halogen HLB VORHAL HID VOLHID1 VOLHID2 VORHID The speed at which participants drove through the sag vertical curves varied. Participants drove two speeds for the large curve and flat roadway sections (45 and 60 mph). Due to the nature of the roadway, the speed at which participants drove through the short curve and left

67 curve was left to each participant’s discretion, and was generally between 25 and 35 mph. Because speed may have an impact on detection distance, vehicle speed was used as the covariate in each of the three analyses. A separate analysis was done for the detection of the speed limit signs because they did not appear in any sag vertical curve. SMART ROAD STUDY RESULTS Analysis of Headlamps Table 30 shows the ANCOVA results of the headlamp analysis for the detection of pedestrians and targets. Significant factors are marked by an asterisk. Table 30. ANCOVA Results for the Headlamp Analysis Source DF Type III SS Mean Square F Value Pr > F Sig Age 1 420875.43 420875.43 13.48 0.0013 * Headlamp 5 1491752.2 298350.44 34.69 <.0001 * Age*Headlamp 5 58050.987 11610.197 1.35 0.2485 Object 1 57821.365 57821.365 24.16 <.0001 * Age*Object 1 234.333 234.333 0.1 0.7572 Curve 3 1673550.1 557850.03 119.7 <.0001 * Age*Curve 3 60794.571 20264.857 4.35 0.0073 * Headlamp*Object 5 126716.48 25343.296 5.12 0.0003 * Age*Headlamp*Object 5 3439.1796 687.8359 0.14 0.9828 Headlamp*Curve 15 534234.23 35615.615 17.14 <.0001 * Age*Headlamp*Curve 15 57150.55 3810.0367 1.83 0.0292 Object*Curve 1 148317.29 148317.29 60.48 <.0001 * Age*Object*Curve 1 6540.3005 6540.3005 2.67 0.1161 Headlamp*Object*Curve 5 131133.58 26226.716 4.99 0.0004 * Age*Headlamp*Object*Curve 5 23842.667 4768.5334 0.91 0.4785 Total 71 4794453.2 * p < 0.05 (significant) Age was found to be a significant factor. The mean detection distance for younger participants (216 ft) was significantly longer than that of older participants (178 ft). This result is expected as visual ability tends to decline with age. Headlamp was also found to be significant. Figure 27 shows the mean detection distance for each headlamp along with the Student-Newman-Keuls (SNK) grouping. The SNK test is a pairwise comparison which looks for a significant difference between each possible pair of factor levels. Factor levels with different SNK groupings (i.e., letters) are significantly different from one another. Factor levels with the same grouping are not significantly different. The HHBs had a significantly longer mean detection distance than all of the low beam headlamps. Of the low beam headlamps, only the VOLHID2s had a significantly different (better) performance.

68 Figure 27. Mean detection distance by headlamp. Object was also found to be a significant factor, with the mean detection distance for targets (216 ft) significantly longer than that of pedestrians (184 ft). This result may have been influenced by the fact that targets were only seen in the flat area and in the large curve, where pedestrians were also seen in the short curve and the left curve. The shorter detection distances in these areas likely brought the mean distance for pedestrians down. Figure 28 shows the significant effect of curve on detection distance. As expected, mean detection distance in the flat area was significantly longer than for any of the sag vertical curves. Of the three curves, the large curve had a significantly longer mean distance than did the short curve which, in turn, had a significantly longer mean distance than did the left curve. Figure 28. Mean detection distance by curve. A B C C C C 0 50 100 150 200 250 300 HHB VOLHID2 HLB VORHID VORHAL VOLHID1 De te ct io n Di st an ce ( ft) Headlamp Mean Detection Distance by Headlamp A B C D 0 50 100 150 200 250 300 Flat Large Curve Short Curve Lef t Curve De te ct io n Di st an ce ( ft) Curve Mean Detection Distance by Curve

69 A significant interaction was found between curve and age. Younger participants had significantly longer detection distances than did older participants on the flat roadway, and in every curve. However, the difference between younger and older participants diminished as the curves got smaller. This is expected as the overall visibility distance is shorter in smaller curves. Figure 29. Mean detection distance by curve and age. A significant interaction was found between headlamp and object. Figure 30 shows that the mean detection distance for targets was significantly higher than that of pedestrians for every headlamp except the HHBs. No difference was found between targets and pedestrians for that headlamp. Figure 30. Mean detection distance by headlamp and object. 0 50 100 150 200 250 300 Flat Large Curve Short Curve Lef t Curve De te ct io n Di st an ce ( ft) Curve Mean Detection Distance by Curve and Age Older Younger 0 50 100 150 200 250 300 350 HHB HLB VOLHID1 VOLHID2 VORHAL VORHID De te ct io n Di st an ce ( ft) Headlamp Mean Detection Distance by Headlamp and Object Pedestrian Target

70 Figure 31 shows the significant interaction of headlamp and curve. The mean detection distance for the flat roadway was significantly longer than for any of the sag curves for every headlamp except for the HHB. For this headlamp – the only high beam lamp – no significant difference was found between the flat roadway and the large curve. In addition, the mean detection distance for the short curve was significantly longer than that for the left curve for every headlamp except the VORHAL. For this headlamp, no significant difference was found between the short curve and left curve. Among the low beam headlamps, the two VOL headlamps had the longest mean detection distances, and the two VOR headlamps had the shortest detection distances in short curves. This is likely due to the difference in light cutoff on the right side of the beam pattern. VOR headlamps have a horizontal cutoff of the headlight on the right side of the beam pattern, where none exists for the HLB and VOL headlamps. In the left curve, there was little difference among headlamps. Figure 31. Mean detection distance by headlamp and curve. A significant interaction was also found between object and curve. No significant difference was found between the mean detection distances for pedestrians and targets on the flat area (242 ft and 246 ft, respectively). In the large curve, however, the mean detection distance for pedestrians (222 ft) was significantly longer than that for targets (186 ft). The reason for this is not immediately clear, but looking at the next significant factor provides some insight. The three-way interaction of curve, headlamp, and object was found to be significant (Figure 32). For the flat area, there was a significant difference between pedestrians and targets for only one headlamp; the HLBs. In the large curve, the mean detection distance for pedestrians was significantly longer for the HHB, HLB, VOLHID1, and VORHID headlamps. However, the greatest difference was for the HHBs. This is likely because the HHB headlamps produce much more uplight than any of the other headlamps. This number is likely responsible for causing the significant difference that was found for the object and curve interaction. In addition, the mean detection distance for pedestrians with the HHB headlamps was significantly higher in the large curve than in the flat area. This is likely due to the fact that, in the curve, the pedestrian was 0 50 100 150 200 250 300 350 400 HHB HLB VOLHID1 VOLHID2 VORHAL VORHID De te ct io n Di st an ce ( ft) Headlamp Mean Detection Distance by Headlamp and Curve Flat Large Curve Short Curve Lef t Curve

71 viewed with a high contrast background (the opposite side of the curved road), whereas the pedestrian had a low contrast background when viewed on the flat road (sky). Figure 32. Mean detection distance by curve, headlamp, and object. Analysis of Headlamp Light Sources and Beam Patterns Additional analyses were done in order to test for significant effects of light source and beam pattern. While there were no significant main effects of either variable, Table 31 shows the ANCOVA results for the significant interactions of light source and curve, and beam pattern and curve. Table 31. ANCOVA Results for the Interaction of Light Source and Curve, and Beam Pattern and Curve Source DF Type III SS Mean Square F Value Pr > F Sig Light Source*Curve 3 32182.505 10727.502 7.9 0.0001 * Beam Pattern*Curve 6 36440.787 6073.4646 4.04 0.0009 * Figure 33 shows that a significant difference between halogen and HID headlamps was only found for the short curve, in which the HID headlamps had a significantly longer mean detection distance. No reason for this effect was immediately apparent, though it was believed that the headlamp beam pattern may have been a confounding variable. The halogen group consisted of one standard-beam headlamp and one VOR headlamp, whereas the HID group consisted of two VOL headlamps and one VOR headlamp. Referring back to Figure 31 in the headlamp analysis, the two VOL headlamps were shown to have the highest mean detection distances in the short curve, other than the HHBs. This led the research team to conclude that the 0 100 200 300 400 500 H H B H LB V O LH ID 1 V O LH ID 2 V O R H A L V O R H ID H H B H LB V O LH ID 1 V O LH ID 2 V O R H A L V O R H ID Flat Large Curve De te ct io n Di st an ce ( ft) Curve and Headlamp Mean Detection Distance by Curve, Headlamp, and Object Pedestrian Target

72 significant interaction found here for light source was most likely due to a confounding effect of beam pattern. Figure 33. Mean detection distance by curve and source. Figure 34 shows the significant interaction of beam pattern and curve. A significant difference between the standard-beam pattern and the VOL pattern was only found in the short curve, in which the VOL headlamps had a significantly higher mean detection distance. This is likely due to the increased uplight on the right side of the VOL beam pattern. In the short curve, the relative position of the pedestrian to the vehicle’s headlamps likely was high enough in the beam pattern to be out of the standard-beam’s hot spot. The standard-beam pattern had significantly longer mean detection distances than did the VOR headlamps in all conditions except in the left curve. This is likely due to a wider hot spot for the VOR headlamps, which reaches further to the left than that of the standard-beam pattern. The VOL headlamps had significantly longer distances than did the VOR headlamps in the large and short curves, but no difference was found in the flat area or left curve. The advantage of the VOL headlamps’ uplight was likely negated in these situations, where the VOR’s light was able to reach further down the road in spite of its horizontal cutoff in the flat area, and where the pedestrian was detected using the left side of the beam pattern in the left curve. Figure 35 shows a comparison of three of the headlamps as examples of the differences among the beam patterns. 0 50 100 150 200 250 Flat Large Curve Short Curve Lef t Curve De te ct io n Di st an ce ( ft) Curve Mean Detection Distance by Curve and Source Halogen HID

73 Figure 34. Mean detection distance by curve and pattern. Figure 35. Comparison of beam patterns. 0 50 100 150 200 250 300 Flat Large Curve Short Curve Lef t Curve De te ct io n Di st an ce ( ft) Curve Mean Detection Distance by Curve and Pattern Standard VOL VOR

74 A separate set of analyses was performed for the detection of speed limit signs, which only appeared on the flat roadway. The only significant factor for all three analyses was age. Younger participants were able to read the signs at a distance of about 395 ft, which was significantly further than the average for older participants (which was approximately 267 ft). The headlamp, light source, and beam pattern had no significant effect on the ability to read speed limit signs. Smart Road Study Discussion The results of this study indicate that beam pattern is an important indicator of visibility in sag vertical curves. It was expected that modern beam patterns (VOL and VOR) would perform worse than a standard-beam pattern in sag vertical curves due to the increased control of uplight, and stricter cutoffs. While the VOR headlamps did perform worse in the flat roadway, the large curve, and short curve, they actually had significantly better performance in the left curve. In addition, the VOL headlamps were found to have no significant difference from the standard headlamp in most conditions, and actually performed better in one (the short curve). Figure 36 shows the change in mean detection distance for the VOL and VOR headlamps as compared to the standard-beam headlamp. These results indicate that a VOL headlamp may actually provide as good or better visibility across all conditions than either a standard beam or VOR headlamp. Figure 36. Change in mean detection distance by curve and beam pattern. Additionally, the differences among beam patterns seem to manifest most strongly in the short curve. However, the VOL headlamps may have had an advantage due to the positioning of the pedestrians in this study. Pedestrians and targets always appeared on the right shoulder of the road which would place them within the portion of the VOL beam pattern which has the most uplight. For objects placed in the roadway, or to the left of the vehicle, the VOL might be found to perform similarly to the other beam patterns. Because beam pattern seemed to be the most important factor for determining visibility in sag curves, one VOL headlamp and one VOR headlamp were selected for use in the follow-up -30 -20 -10 0 10 20 30 Flat Large Curve Short Curve Lef t Curve Di st an ce (f t) Curve Change in Mean Detection Distance by Curve and Beam Pattern Standard VOL VOR

75 Public Road study. Specifically the VOLHID2 and VORHID headlamps were selected so that a direct comparison of beam pattern could be made without the potential confound of light source. PUBLIC ROAD EXPERIMENTAL DESIGN The second phase of the experiment took place on public roads in Blacksburg and Christiansburg, VA, and compared VOL and VOR headlamps across 11 sag vertical curves of varying sizes. The study used a 2 (Headlamp) x 12 (Sag Vertical Curve) full-factorial design. Independent Variables Several independent variables were manipulated or controlled for this experiment. They are listed below. Between-Subjects Variables • Gender (2 levels): Female, Male. The gender independent variable was chosen in order to generalize the results of this study to a broad user population. This factor was used for balance only; it was not used in the data analysis. • Age (2 levels): Younger (21-34), Older (65+). The younger and older age groups were selected to investigate the changes in vision and perception that may occur with increasing age. Within-Subjects Variables • Headlamp (2 levels): VOL, VOR. The two sets of headlamps were chosen in order to compare the performance of the two low-glare beam patterns in sag vertical curves. • Sag Vertical Curve (12 levels): Eleven sag vertical curves and two flat areas were utilized for the study, creating 12 different levels. Sag vertical curves were given a designation based on which road they were on, which is described in more detail in the Facilities and Equipment section below. The sag curves selected encompass many different lengths across three roadway types (divided highway, two-lane highway, and residential). Dependent Variables Detection Distance The distance at which a participant could identify a target was recorded as a measure of visibility for each curve and flat area. Participants were instructed to verbally identify targets as they drove by saying the word “target.” At that moment, an in-vehicle experimenter would flag the data with a button press. When the vehicle reached the target, the in-vehicle experimenter would again flag the data with a button press. Later analysis determined the distance traveled between these two points, and this was termed the “detection distance.” Figure 37 illustrates the two points at which the data was flagged. Detection distance as defined in this report may be more closely related to recognition distance in that participants had to identify what an object was, rather than simply detect that something was there. However, as there was only one object used in this study, it is likely that detection and recognition were nearly simultaneous.

76 Figure 37. Determining detection distance. Participants Twenty-four participants were selected to take part in this study. Participants were selected from two age categories: younger (18-34 years old) and older (65+). Six younger males, six older males, six younger females, and six older females participated. Recruitment occurred through the VTTI participant database and word-of-mouth. A general description of the study was provided to the subjects over the phone before they decided if they were willing to participate. If they were interested, subjects were then screened with a verbal questionnaire to determine whether they were licensed drivers and whether they had any health concerns that should exclude them from participating in the study. If subjects were determined to be eligible for the study, they were then scheduled to come to VTTI for participation. When subjects arrived at VTTI, they read and signed an informed consent form. Subjects were paid $20/hr and were allowed to withdraw at any point in time, with compensation adjusted accordingly. Participants’ Age and Visual Capabilities Younger participants’ ages ranged from 21 to 32 years old, with a median age of 23 years. Older participants’ ages ranged from 65 to 69 years old, with a median age of 67 years. One participant failed a color-blindness test. All other participants passed, with only two participants giving a single incorrect response each. Table 32 presents the distribution of visual acuity scores. A minimum of 20/40 was required for participation. Participants were allowed to wear corrective lenses if they indicated that they normally wear them while driving.

77 Table 32. Visual Acuity Scores Visual Acuity (Snellen) Number of Participants 20/13 4 20/15 3 20/20 8 20/25 5 20/30 3 20/40 1 Participants’ visual acuity was also measured in the presence of glare using a BAT. The distribution of visual acuity scores for each eye and level of glare are presented in Table 33. Table 33. Visual Acuity in the Presence of Glare No Glare Low Glare Med Glare High Glare Acuity Score Left Eye Right Eye Left Eye Right Eye Left Eye Right Eye Left Eye Right Eye 20/13 2 3 3 2 1 1 1 1 20/15 3 2 1 1 2 1 0 0 20/20 7 5 3 6 5 6 3 4 20/25 6 6 8 6 5 5 5 4 20/30 2 5 5 3 5 5 5 5 20/40 2 1 1 5 3 1 5 5 20/50 1 2 1 0 1 3 4 2 20/60 and worse 1 0 2 1 2 2 1 3 Facilities and Equipment Test Route This experiment was conducted on public roads in Blacksburg and Christiansburg, VA. The route encompassed 11 sag vertical curves on three types of roadways: divided highway (four curves), two-lane highway (five curves), and residential (two curves). Figure 38 shows a map of the route. Participants departed from VTTI, drove to a cul-de-sac in a residential neighborhood where they turned around, and then drove to the end of the route marked by the number 3 on the map. At this point participants switched vehicles, and returned to VTTI driving the same route in the opposite direction.

78 Figure 38. Test route (Source: Google). Measurements of a majority of the sag vertical curves were found using the design documents which were supplied by the Virginia Department of Transportation (VDOT). In cases where the design documents were unavailable, measurements were made using GPS data recorded using a GPS-enabled vehicle. The data were processed in ArcGIS™ and ArcMap™ 10 software to generate the elevation, giving a profile of the curves. The tangent lines were found visually using a straight edge. The slopes of the tangent lines were calculated to determine the algebraic change in grade while the horizontal distance between the tangent points was used to determine the length of the curves. Table 34 summarizes the curve information. It also shows the minimum K value based on design speed, and the actual K value – called K reality or Kr – based on the design documents or the grade and length measurements. Shaded cells indicate curves which were not designed to the AASHTO criteria. Radford Road was designed in 1940, before the first AASHTO Green Guide was issued. Route 460 was designed in 1965 after the AASHTO guide was introduced, but the mountainous terrain may have prevented designing to these standards.

79 Table 34. Sag Vertical Curve Measurements Road Name Road Type Design Speed (mph) Minimum K Algebraic Change in Grade, A (%) Length, L (ft) KR (L/A) Route 460 East Divided Highway 60 136 4.73 600 127 Route 460 East Divided Highway 60 136 12.00 800 67 Route 460 West Divided Highway 60 136 4.28 600 140 Route 460 West Divided Highway 60 136 12.00 800 67 Independence Blvd* Residential 25 26 18.75 500 27 Radford Rd Two-lane Highway 45 79 10.00 500 50 Radford Rd Two-lane Highway 45 79 6.00 500 83 Radford Rd Two-lane Highway 45 79 5.90 500 85 Radford Rd Two-lane Highway 45 79 4.14 500 121 Radford Rd Two-lane Highway 45 79 9.22 500 54 Sapphire Ave* Residential 25 26 19.17 320 17 * Measurements for these curves were made by the research team as the design documents were unavailable. Headlamps and Test Vehicles Participants drove a 1999 and 2000 model Ford Explorer, each equipped with a different set of headlamps. The headlamps used in the study are summarized in Table 35 below. These headlamps were selected to investigate the effect of beam pattern on visibility in sag vertical curves.

80 Table 35. Headlamps Headlamp Abbreviation Description Original Vehicle Photo VOL Visually Optically Aligned Left HID Mercedes E320 VOR Visually Optically Aligned Right HID Lincoln Navigator The test vehicles were also equipped with DASs. The DAS recorded video both inside and outside of the vehicle and button inputs from the in-vehicle experimenter, as well as network data such as speed and distance. Detection distance was determined by analysis of this data. Public Road Experimental Method Participants were initially contacted and screened on the phone using an internal VTTI database of persons who had expressed interest in participating in research studies. When participants arrived at VTTI for participation, they first read and signed the informed consent. Participants then filled out a W9 tax form, and a health history questionnaire. These were followed by vision tests for acuity, contrast sensitivity, and color blindness. Participants were only excluded from participation if visual acuity was less than 20/40 (the legal minimum to hold a Virginia driver’s license), or if they had taken any substance which might impair their ability to drive. Participants were scheduled in pairs. Once the paperwork and vision tests were complete, an in-vehicle experimenter explained the instructions for the study and answered any questions the participants had. The participants were then escorted to their assigned vehicle by an in-vehicle experimenter. The participant in the first vehicle was instructed to begin driving the route, while the second participant waited approximately 60 seconds before beginning the drive. This kept the two vehicles apart during the drive so that one would not interfere with the other. Participants were given turn-by-turn directions by the in-vehicle experimenter so that they did not have to memorize the route. At the end of the route, participants were instructed to stop in a parking lot. Here, the participants switched vehicles before retracing the route back to VTTI. As participants drove, they would verbally identify the targets as they encountered them. Targets always appeared on the driver’s right hand side. In addition to identifying targets,

81 participants read aloud the number on any speed limit sign they encountered as soon as they could read it. This task was used only to keep participants active and alert during long stretches between sag vertical curves. The point at which participants identified a target was flagged in the data by the in-vehicle experimenter by pressing a button. The moment at which the vehicle reached the target was also flagged by another button press. Once participants had completed the route and returned to VTTI they were paid for their participation, and given a receipt and a copy of the informed consent. Participants were compensated at a rate of $20 per hour. Presentation Orders Because all participants drove the same route, the presentation order of the sag vertical curves was fixed. The order of the headlamps was balanced with half of the participants using VOL headlamps first, followed by the VOR headlamps, and the other half of participants using VOR headlamps first, followed by the VOL headlamps. Table 36 shows which curves were seen with which headlamps for a pair of participants. Directions 1 and 2 refer to the first and second half of the route. All participants saw the curves in the residential area (IND and SAP) in both directions with both headlamps. Table 36. Headlamp Presentation Orders Participant 1 Participant 2 Curve Direction 1 Direction 2 Direction 1 Direction 2 460E1 VOL VOR 460E2 VOL VOR IND VOL/VOR VOL/VOR VOL/VOR VOL/VOR SAP VOL/VOR VOL/VOR VOL/VOR VOL/VOR RAD1 VOL VOR VOR VOL RAD2 VOL VOR VOR VOL RAD3 VOL VOR VOR VOL RAD4 VOL VOR RAD5 VOR VOL Flat VOL VOR VOR VOL 460W1 VOR VOL 460W2 VOR VOL Public Road Data Analysis An ANCOVA with a significance level of 95% (α = 0.05) was used. Because the speed limits on the sag vertical curves varied, and speed may have an impact on object detection, vehicle speed was used as the covariate. PUBLIC ROAD STUDY RESULTS Table 37 shows the ANCOVA results for the detection of targets. No factors were found to be significant.

82 Table 37. ANCOVA Results for the Headlamp Analysis Source DF Type III SS Mean Square F Value Pr > F Sig Age 1 2940.6471 2940.6471 0.27 0.6108 Headlamp 1 11246.658 11246.658 3.51 0.0766 Age*Headlamp 1 159.81033 159.81033 0.05 0.8257 Curve 11 53046.425 4822.4023 1.73 0.0683 Age*Curve 11 41229.965 3748.1787 1.35 0.2006 Headlamp*Curve 5 13482.899 2696.5798 1.58 0.1772 Age*Headlamp*Curve 5 5007.1958 1001.4392 0.59 0.7099 Total 35 127113.6 * p < 0.05 (significant) While no factors were found to be significant, headlamp and curve both had relatively low p values. (0.07 and 0.06, respectively). For the headlamp factor, the VOL headlamps had a mean detection distance of 91ft compared to 78 ft for the VOR headlamps. This is a small difference, but it does appear that the VOL headlamps provide slightly better visibility than VOR headlamps, at least for objects on the right side of the vehicle. While curve was not found to be significant, SNK pairwise comparisons show that there were significant differences between curves. Figure 39 shows the mean detection distance for each curve as well as the flat area. Some curves had longer mean detection distances than the flat area, and even though they were not significantly different, it seems counterintuitive. A possible factor in this may have been the presence of other vehicles. The section of Route 460 used in the study is a particularly busy section, and it is possible that the headlights from a leading vehicle illuminated the targets for a participant. Another interesting result is that the two curves on Route 460 Westbound had significantly higher detection distances than the virtually identical curves on Route 460 Eastbound. This is likely the result of an order effect. The two targets on the eastbound lanes were the first that participants encountered, and the two targets on the westbound lanes were the last two that participants encountered. It is likely that participants improved at detecting targets as the night went on. That is to say that they became more aware of what they should be looking for, and where the targets might appear.

83 Figure 39. Mean detection distance by curve. Public Road Study Discussion Interestingly, only the two curves in the residential neighborhood (IND and SAP) had a mean detection distance that was significantly shorter than the flat area. All other curves were not statistically different from the flat detection distance. The reason that the two residential curves were the only curves to have significantly shorter detection distances than the flat roadway was not immediately apparent. One direction of the Sapphire Avenue curve (SAP) had a crest vertical curve just prior to the sag curve which blocked the view of the target at a certain distance, but this did not appear to be an issue because the standard deviation for that curve was quite small (5.25 ft for VOL headlamps, and 2.5 ft for VOR headlamps). The mean detection distance for each curve was plotted against each measured aspect of the curve (length, change in grade, KR) to determine if a relationship could be found to explain at what point a curve’s mean detection distance could become diminished as compared to flat detection. Figure 40 shows the mean detection distance by curve length, with the two residential curves (IND and SAP) indicated by the two white squares. The figure also shows the linear trend line with the associated R2 value. While the shortest curve did have the lowest mean detection distance, there is only a weak relationship between detection distance and curve length. There were numerous curves with lengths similar to the IND curve which were not significantly different from the flat roadway. Thus, the point at which a curve’s detection distance becomes significantly less than flat roadway cannot be explained solely by curve length. A A AB AB AB AB B B B BC C C 0 20 40 60 80 100 120 140 160 De te ct io n Di st an ce ( ft) Curve Mean Detection Distance by Curve

84 Figure 40. Mean detection distance by curve length. Figure 41 shows the mean detection distance by the algebraic change in grade for each curve, along with the linear trend line and associated R2 value. There is a weak negative correlation between A and detection distance, and the two residential curves which had significantly shorter mean detection distances than the flat roadway are shown to have the largest values for A by far. This may indicate that the point at which a curve’s detection distance becomes less than flat roadway may be when that curve’s algebraic change in grade is somewhere between 13% and 18%. Figure 41. Mean detection distance by algebraic change in grade. Finally, Figure 42 shows the relationship between detection distance and the rate of curvature, KR, as well as the linear trend line and associated R2 value. A weak positive R² = 0.306 0 50 100 150 200 0 200 400 600 800 1000 De te ct io n Di st an ce ( ft) L (ft) Mean Detection Distance by Curve Length, L Not Signif icantly Dif ferent f rom Flat Signif icantly Shorter than Flat R² = 0.4075 0 50 100 150 200 0 5 10 15 20 25 De te ct io n Di st an ce ( ft) A (%) Mean Detection Distance by Algebraic Difference in Grade, A Not Signif icantly Dif ferent f rom Flat Signif icantly Shorter than Flat

85 correlation was found where a larger K value tends to result in longer detection distances. The two residential curves which had mean detection distances which were significantly shorter than flat roadway had the lowest KR values. Thus, the point at which a curve’s detection distance becomes less than flat roadway may be somewhere between KR values of 27 and 50. Figure 42. Mean detection distance by KR. A much stronger relationship between KR and detection distance was expected. However, while the relationships between detection distance and the various curve measures were weak, there did appear to be some thresholds for the change in grade, A, and rate of curvature, KR, that might explain which curves will have reduced visibility as compared to flat roadway. Unfortunately, those thresholds could not be pinpointed within this data. Further research could attempt to validate these findings, and narrow the window by examining visibility in curves with a wide range of A and KR values. R² = 0.3705 0 50 100 150 200 0 50 100 150 De te ct io n Di st an ce ( ft) KR (L/A) Mean Detection Distance by the Rate of Curvature, KR Not Signif icantly Dif ferent f rom Flat Signif icantly Shorter than Flat