Solid-State Roadway Lighting Design Guide: Volume 2: Research Overview (2020)

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.

38 According to the project plan, the tasks to be completed as part of this project included the execution of the research work plan (Smart Road testing), development of the SSL Guide, and inclusion of a research and implementation roadmap of recommended future research and the approximate associated costs. The project plan also includes ongoing monitoring of research currently being conducted in areas of relevance to this report and the SSL Guide. These areas include • Light and human health, • Light and environmental impacts, • Light and adverse weather, • Light and aging, • Lighting design metrics, and • Comparison of crashes at HID and SSL installations. Any new data that emerged before the completion of the final document have been included. Smart Road Testing An experiment on the closed Smart Road test track at VTTI was undertaken to identify critical metrics that can be used to determine the quality and usefulness of a roadway lighting system. This experiment was a driving task in the lighted test environment where the perfor- mance of the critical visual task by drivers was investigated in terms of the lighting design. This critical visual task was determined as a result of the assessment of safety performed as part of this project. Experimental Methods Participants Participants were recruited from VTTI’s internal database. In addition, printed flyers and e-mail listserv announcements were used to recruit participants. Participants were recruited from two age ranges: 18 to 35 years (younger) and 65 years and older (older). Sixty participants were recruited to participate in the study. The participant sample was age balanced to have an equal number of younger and older participants. Participants had a valid U.S. drivers’ license, had a visual acuity of at least 20/40 (measured by the Early Treatment Diabetic Retinopathy Study chart with an illuminator cabinet), and were not color blind (measured by the Ishihara Color Vision Test). C H A P T E R 5 Research and Testing Results

Research and Testing Results 39 Experimental Design A repeated-measures experimental design was used to assess the effects of light SPD, light level, surround ratio, and uniformity ratio on driver visual performance. Driver visual performance was measured according to the distance at which participants detected objects as they drove through the test environment. Pedestrians and small targets were used for the detection task. The independent variables used in this experiment are summarized in Table 4. Independent Variables As seen in the experimental design, several variables were controlled or manipulated during the study. Because this study used human participants, the limiting duration of each experi- mental session was critical to reducing the potential for confounding effects of learning and fatigue, which were important in this experiment because of the higher number of independent variables. Some of the independent variables were classified as between-subjects, while the remaining were classified as within-subjects. Between-subjects variables are those that cannot be manipulated within an experimental session, such as the SPD of the light and the surround ratio. In addition to those two variables, the independent variables included the uniformity ratio, age, light level, speed, object, object color, and object offset. Spectral Power Distribution of the Light. Three commercially available LED luminaire types with varying SPDs and an HPS light source were tested (Figure 16). In the absence of an established metric for distinguishing the SPDs of luminaires, CCT was used to evaluate the effect of SPDs on visual performance. Three distinct CCTs—3000K LED, 4000K LED, and 5000K LED—were selected for the purpose of the evaluation. Luminaires of 3000K LED and 5000K LED CCT represented the minimum and the maximum range of color temperatures currently available in the market; 4000K LED is widely used for roadway lighting on the basis of the results of the survey discussed in Chapter 2. As a reference, an HPS light source was used at a single light level. The LED luminaires used for the study were obtained from a single manufacturer and had matching light intensity distributions. Dimming was used to match the lighting levels. All the luminaires used for this study were Type-II medium-throw luminaires. Independent Variable Level Classification Light SPD 3000K LED 4000K LED 5000K LED HPSa Between subjects Light level (cd/m2) High (1.5) Medium (1.0) Low (0.7) Within subjects Surround ratio (average shoulder illuminance to average lane illuminance) High (0.8) Low (0.45) Between subjects Uniformity ratio (average luminance to minimum luminance) High (1.8–3.5) Low (1.3–1.4) Between subjects Speed (mph) High (55) Low (35) Within subjects Age (years) Old (≥65) Young (18–35) Between subjects aOnly at medium light level. Table 4. Independent variables and their levels used in the study.

40 Solid-State Roadway Lighting Design Surround Ratio. The surround ratio is the measure of how bright the roadway travel lanes are with respect to the adjacent areas of paved and unpaved shoulder. CIE 140-2000 defines surround ratio as the average horizontal illuminance on the two longitudinal strips, each adja- cent to the two edges of the carriageway, to the average horizontal illuminance on two longi- tudinal strips, each adjacent to the two edges of the carriageway, but lying on the carriageway. Two surround ratios (Table 4) were tested to examine the impact of surround ratio on visual performance. Uniformity Ratio. The uniformity ratio is defined as the ratio of average to minimum pave- ment luminance of the roadway. Two levels of uniformity (Table 4) were assessed to examine how the uniformity ratio affects the visibility of objects. Age. Participants were divided into two age groups (Table 4) to examine how physio logical changes in the eyes affect the visual performance of drivers. Younger drivers typically have a higher-performing visual system that can be partially offset by the years of driving experience. Including age allows a wide range of physiological capabilities and driving experiences to be considered. Light Level. The light level is the average pavement luminance level of roadway. Three light levels were used in the study (Table 4). This factor was included to examine the effect of lighting levels on visual performance. As noted above, the light fixtures had similar intensity distribu- tions, and a dimming system was used to control the light level of the roadway. These levels were verified through a photometric analysis of each lighting condition. The horizontal and vertical illuminance were also characterized for all lighting levels in the roadway and on the shoulder of the road. Speed. Because speed can affect the field of view of the driver, two speeds were selected to represent common driving scenarios: highway at 55 miles per hour (mph) and arterial at 35 mph. Object. Two objects of different sizes and shapes were selected to examine how different lighting configurations affect the visibility of different objects. Small targets represented hazards 5000K LED 4000K LED 3000K LED 2100K HPS Figure 16. Spectral power distributions of the light sources used to measure drivers’ visual performance and perception of discomfort glare.

Research and Testing Results 41 that may be on or near the roadway and are just large enough that they may cause damage to a vehicle, and pedestrians were used because they are commonly encountered in many driving scenarios and can also represent the presence of wildlife. Each of these objects was presented in a different color and at a different offset distance from the roadway to simulate hazards on the roadways. Object Color. Roadway lighting with different SPD values can affect the perception of the color of objects on or near the roadway. For targets, three color variants (red, blue, and gray) were used to determine whether the SPD of the lighting affects the visibility differently. For pedestrian, four color variants (red, blue, gray, and black) were used to determine whether the SPD has an effect on the visibility of pedestrians’ clothing color. Object Offset. Objects and pedestrians may appear at different distances from the driver’s lane and will therefore be lit differently on the basis of their location within the intensity distri- bution of the roadway luminaires. Three different distances from the roadway were used in the experiment. For targets, two offsets were used: 2 ft left of the driving lane (indicated by 2L in Figure 17) and 2 ft right of the driving lane (indicated by 2R in Figure 17). For pedestrians, the two offsets used were 2 ft right of the driving lane (indicated by 2R in Figure 17) and 10 ft right of the driving lane (indicated by 10R in Figure 17). The vertical illuminance levels on the pedes- trian locations and the target locations were matched across the lighting conditions (Table 5 and Table 6). Dependent Variables Detection Distance. Detection distance was the dependent variable in the experiment. Detection distance is the distance between the object that is being detected and the driver when the driver detects the object. Detection distance is a commonly used measure of visual perfor- mance in nighttime roadway visibility studies (Bhagavathula and Gibbons, 2013; Bhagavathula et al., 2018; Mayeur et al. 2010; Shinar 1985). Discomfort Glare Rating. Discomfort glare was measured by using a rating scale, as shown in Table 7. This scale has been reported to produce reliable data, with smaller numbers meaning lower discomfort glare and higher numbers meaning higher discomfort glare (Bhagavathula and Gibbons 2018; Fisher 1991; Tyukhova 2015; Tyukhova and Waters 2018). The scale also has an anchor of zero for no discomfort glare. Equipment and Facilities Virginia Smart Road This study was conducted on the Virginia Smart Road, which is a 2.2-mile-long, controlled- access test track. The Smart Road is equipped with a configurable roadway lighting system that Figure 17. Locations of offset locations with respect to vehicle travel direction on the Virginia Smart Road.

42 Solid-State Roadway Lighting Design Light Type Pedestrian Offset Target Offset 2 ft Right 10 ft Right 2 ft Right 2 ft Left High Light Level 2100K HPS 19.9 14.9 17.5 21.1 3000K LED 21.5 15.1 18.5 21.9 4000K LED 20.1 15.9 17.8 22.7 5000K LED 22.3 16.7 18.3 22.5 Medium Light Level 2100K HPS 13.8 9.7 11.1 13.5 3000K LED 14.7 10.6 11.8 14.2 4000K LED 14.3 9.2 12.4 13.6 5000K LED 13.7 10.2 11.4 14.4 Low Light Level 2100K HPS 8.9 5.8 7.1 8.9 3000K LED 9.4 6.5 6.9 9.1 4000K LED 9.1 5.9 7.6 9.6 5000K LED 8.5 6.3 7.4 9.7 Note: HUR = high uniformity ratio; HSR = high surround ratio. Table 5. Vertical illuminance measured at the pedestrian and target locations in the HUR-HSR shield condition. Shield Condition Pedestrian Offset Target Offset 2 ft Right 10 ft Right 2 ft Right 2 ft Left High Light Level HUR-LSR 19.6 12.6 16.8 21.6 LUR-HSR 20.9 13.6 16.6 20.4 LUR-LSR 19.0 12.5 16.2 20.0 Medium Light Level HUR-LSR 10.3 8.5 10.3 13.3 LUR-HSR 11.8 9.4 10.1 13.2 LUR-LSR 9.8 7.8 9.9 13.6 Low Light Level HUR-LSR 7.9 7.1 6.2 9.9 LUR-HSR 7.5 7.0 6.8 10.9 LUR-LSR 7.4 5.5 6.2 10.0 Note: LSR = low surround ratio; LUR = low uniformity ratio. Table 6. Vertical illuminances measured at the pedestrian and target locations in the surround and uniformity shield conditions.

Research and Testing Results 43 includes 75 poles that each support three luminaires (Figure 18). The lighting system can be configured to 40 m, 80 m, or 120 m spacing between lights. The lighting system can be controlled remotely, which allows researchers to change luminaires, spacing, or light levels as needed. For the current study, a spacing of 262 ft (80 m) and a mounting height of 49 ft (15 m) were used. To manipulate the surround and uniformity ratios, custom shields were designed and fitted to the luminaires (Figure 19). Three sets of shields were specifically designed to manipulate the surround and uniformity ratios. One shield increased the uniformity ratio (the road was not evenly lighted). In the second shield condition, the surround ratio was increased (light levels on areas adjacent to the travel lane—for example, the shoulders—were lowered). In the final shield condition, both uniformity and the surround ratios were increased. The shields were designed to be clipped on to the luminaires. Test Vehicles The test vehicles were two identical 2017 Ford Explorers. Each vehicle was equipped with a data acquisition system that captured four camera views inside and outside the vehicle, the Global Positioning System (GPS), and vehicle network data. Targets Targets were 7-in.2 pieces of plywood painted with coarse, nonreflective paint (Figure 20). Targets were propped up with wooden stands and placed approximately 2 ft outside the driving lane on either the right or left side of the participant’s driving lane. For safety, targets were designed to break if struck by the participant vehicle. The spectral reflectance of the targets is shown in Figure 21. The colors of the targets were selected because they represent two opposite ends of the color spectrum and a neutral color. As a result of this color choice, the benefits of the high and the low CCT light sources would be highlighted. Description Rating No discomfort glare 0 Glare between nonexistent and noticeable 1 Glare noticeable 2 Glare between noticeable and disagreeable 3 Glare disagreeable 4 Glare between disagreeable and intolerable 5 Glare intolerable 6 Table 7. Discomfort glare rating scale used in this study. Figure 18. Smart Road lighting system.

44 Solid-State Roadway Lighting Design Figure 19. Custom shields: (a) no shield on the luminaire, representing high uniformity–high surround; (b) luminaire with high uniformity–low surround shield; (c) luminaire with low uniformity–high surround shield; and (d) luminaire with low uniformity–low surround shield. (a) (b) (c) Figure 20. Targets: (a) red, (b) gray, and (c) blue. Red target Gray target Blue target Figure 21. Spectral reflectance of the three colored targets used in the study.

Research and Testing Results 45 Pedestrians The pedestrians were on-road experimenters wearing colored surgical scrubs (Figure 22), which allowed the pedestrian to have a consistent color contrast. Four colored scrubs were used: red, black, gray, and blue. Pedestrians always stood facing the roadway with their arms by their side, so that they appeared similarly across lighting configurations. The spectral reflectance of these targets is shown in Figure 23. The color selection was the same as for the targets, but an additional low luminance neutral was added for further analysis possibilities. Characterization of Illuminance The horizontal and vertical illuminance was characterized for all types of luminaires without shields (3000K LED, 4000K LED, 5000K LED, and 2100K HPS) and for one light type (4000K LED) with all of the shield conditions (HUR-LSR, LUR-HSR, and LUR-LSR) at all the light levels. All the luminaires were of the same type and were from the same manufacturer, so it was assumed that the light levels would not change for the luminaires of other CCTs. The illuminance was characterized by means of the trailer-based Roadway Lighting Mobile Measurement System. Data were collected by horizontal illuminance sensors at the quarter-lane points. The data capture rate was 20 Hz, and the vehicle travel rate was for a spatial sample rate of approximately 0.9 m. In addition, vertical illuminance was captured opposite the direction of Figure 22. Pedestrians in colored scrubs. Blue scrubs Red scrubs Gray scrubs Black scrubs Figure 23. Spectral reflectance of the colored scrubs used in the study.

46 Solid-State Roadway Lighting Design travel with two illuminance meters mounted at 5 ft (∼1.5 m) high and at the quarter-lane posi- tion. Summarized measurements of average illuminance, uniformity, and surround ratios are shown in Table 8, Table 9, and Table 10, respectively. Detailed illuminance measurements are provided in Appendix C. The vertical illuminance on the pedestrians and targets was measured by an illuminance meter (Minolta T-10A). For the pedestrians, the illuminance was measured at 5 ft (∼1.5 m) from the road surface facing the direction of vehicle. For the targets, the vertical illuminance was mea- sured by placing the illuminance flush with the side facing the vehicle when the target was placed on the road. The vertical illuminance values for pedestrians and target locations are shown in Table 5 and Table 6. Experimental Procedure Participant Recruitment, Consent, and Compensation Participants were recruited from VTTI’s internal database of volunteers. Participants were contacted by telephone to determine whether they were interested and eligible for participa- tion; if so, they were scheduled for an experimental session. When participants arrived at VTTI, they filled out an informed consent form and performed some simple vision tests to ensure Light Type Shield Light Level Average Horizontal Illuminance (lux) Average Vertical Illuminance (lux) 2100K HPS HSR-HUR Medium 12.3 10.5 5000K LED High 20.2 15.9 4000K LED 19.0 16.3 3000K LED 18.9 16.2 5000K LED Low 8.9 8.7 4000K LED 9.4 8.6 3000K LED 9.9 7.9 5000K LED Medium 12.9 10.5 4000K LED 11.5 10.2 3000K LED 12.6 10.1 4000K LED HUR-LSR High 17.7 10.5 Low 9.6 6.1 Medium 11.0 7.1 LUR-HSR High 17.2 13.2 Low 8.5 6.6 Medium 10.8 8.3 LUR-LSR High 17.3 13.9 Low 8.3 6.4 Medium 12.7 9.4 Table 8. Average vertical and horizontal illuminance levels across different light types, shield conditions, and light levels.

Research and Testing Results 47 Light Type Shield Light Level Uniformity Ratio (Average/Minimum Illuminance) 2100K HPS HSR-HUR Medium 7.5 5000K LED High 7.5 4000K LED 3.8 3000K LED 3.8 5000K LED Low 4.9 4000K LED 3.7 3000K LED 5.8 5000K LED Medium 6.1 4000K LED 4.6 3000K LED 6.8 4000K LED HUR-LSR High 4.1 Low 7.2 Medium 5.1 LUR-HSR High 6.7 Low 4.3 Medium 4.6 LUR-LSR High 3.5 Low 4.5 Medium 3.6 Table 9. Uniformity ratios of illuminance across different light types, shield conditions, and light levels. they were eligible. Eligible participants performed the experimental tasks described below. Participants were paid $30 per hour of participation. Participants were recruited for two sessions and experienced the same lighting conditions at the two different speeds in each session. Procedures Two participants were scheduled for each experimental session. Once the paperwork and vision screening were completed, participants were read a script that described the experimental tasks. Participants were also shown the discomfort rating scale and were given a definition of discomfort glare and how to provide the discomfort glare rating. Participants were then escorted to the two experimental vehicles. Each participant drove one of the vehicles, and in-vehicle experimenters accompanied each participant. Each participant drove laps on the Smart Road. Only one participant drove through the lighted section of the Smart Road at a time, so as to allow both participants to drive laps without ever driving past the other and without interference from the other vehicle’s headlamps. Participants drove a total of 12 laps. After every four laps, the light levels on the road were changed. At the end of first, fifth, and ninth laps, the in-vehicle experimenter asked the partici- pants to rate the amount of discomfort glare according to the scale. Copies of the scale were provided to the participants for reference.

48 Solid-State Roadway Lighting Design As participants drove through the laps, several objects were presented at designated stations. Participants were asked to verbally identify the type of object (pedestrian or target) as soon as it was visible. The in-vehicle experimenter flagged these moments in the data stream by pressing a hand-held button. The GPS coordinates of the object’s locations were predetermined. The GPS coordinates at detection and recognition were cleaned up at a later time. The detections were adjusted to the point in time at which the participants said “target” or “pedestrian,” thereby eliminating the time delay resulting from experimenter input. A high-precision differential GPS unit was used in the experimental vehicle and to collect the GPS locations of the worker. The GPS system had an accuracy of about 0.33 ft (0.1 m). The presentation of the objects, speeds, and light levels was counterbalanced across several orders to offset any learning effects. Once all 12 laps were complete, participants were instructed to exit the Smart Road and return to VTTI. There, they were compensated and released. The same process was repeated for the second experimental session at a different speed. Analyses Visual Performance Across LED Light Sources The goal of this analysis was to determine whether there were differences in visual perfor- mance between LED light sources of different CCTs across different light levels, surround ratios, and uniformity ratios. Two analyses were conducted to assess visual performance differences across LED light sources. Light Type Shield Light Level Surround Ratio (Average Shoulder/Average Lane Illuminance) 2100K HPS HSR-HUR Medium 0.9 5000K LED High 0.9 4000K LED 0.8 3000K LED 0.8 5000K LED Low 0.9 4000K LED 0.9 3000K LED 0.7 5000K LED 0.8 4000K LED Medium 0.8 3000K LED 0.9 4000K LED HUR-LSR High 0.5 4000K LED Low 0.5 4000K LED Medium 0.5 4000K LED LUR-HSR High 0.8 4000K LED Low 0.7 4000K LED Medium 0.7 4000K LED LUR-LSR High 0.6 4000K LED Low 0.5 4000K LED Medium 0.6 Table 10. Surround ratios of illuminance across different light types, shield conditions, and light levels.

Research and Testing Results 49 Detection Distance Analysis. In the first analysis, a linear mixed model (LMM) was used to compare detection distances across all the light sources. Two LMM analyses were used to assess the effects of the type of light source, the light level, the surround ratio, and the unifor- mity ratio on detection distance. One LMM was used for the pedestrian detection task and the other was used for the target detection task. Safe Stop Analysis. In the second analysis, each participants’ detection distance under each condition was classified as safe or not safe. The analysis was termed “safe stop analysis.” A detec- tion was classified as safe if the detection distance was greater than the stopping sight distance (SSD) for the speed at which the vehicle was traveling during the detection task. AASHTO uses SSD as a design standard for sight distance in road design (AASHTO 2018a); it is the minimum distance required for a driver traveling at the design speed to come to a complete stop. If the detection distance is lower than the SSD, then it can be assumed that, under the given conditions, the driver would not be able to come to a safe stop; therefore, that particular distance would be classified as not safe. For the second analysis, a mixed models logistic regression (MMLR) was performed with the type of stop (safe versus not safe) as the dependent measure. Two MMLR analyses were used to assess the effects of the type of light source, the light level, the surround ratio, and the uniformity ratio on the types of safe versus not safe stops. One MMLR was used for the pedestrian detection task and the other for the target detection task. MMLR allows the estimate of a specific factor to be considered as an averaged value over all the levels of other factors included in the same model. Thus, the confounding effect can be effectively addressed by simply including multiple factors that might simultaneously confound each other in the model. Visual Performance Across LED, HPS, and Unlighted Conditions The goal of this analysis was to determine whether there were differences in visual perfor- mance between LED light sources of different CCTs, a legacy HPS light source, and unlighted conditions. As with the visual performance analysis across LED light sources, a two-pronged approach was used for this analysis. In the first analysis, LMMs were used to assess the differ- ences in detection distance between LED, HPS, and unlighted conditions. In the second analysis, MMLRs were used to assess the differences in the type of stop between LED, HPS, and unlighted conditions. Two LMMs and MMLRs were used to assess the effects of the type of light source on detection distance and the type of stop (one for the pedestrian detection task and the other for the target detection task). Visual Performance Across Uniformity Ratios of LED and HPS Light Sources Uniformity ratios varied across unshielded (high uniformity and high surround ratios) LED light sources and HPS light sources (1.3 versus 3.5). As a result, it was anticipated that, for achromatic detections of pedestrian and targets, the difference in the uniformity ratios between the light sources, rather than the SPD, could affect visual performance. Two LMMs were con- ducted at the medium light level for both targets (gray colored) and pedestrian (gray clothing) detection tasks with the detection distance as the dependent measure and the uniformity ratio, speed, and offset as independent measures. Perception of Discomfort Glare Across LED Light Sources The goal of this analysis was to determine whether differences in perception of discomfort glare occurred between LED light sources of different CCTs across difference light levels, surround ratios, and uniformity ratios. One LMM was used to assess the effects of the type of light source, the light level, the surround ratio, and the uniformity ratio on the perception of discomfort glare.

50 Solid-State Roadway Lighting Design Perception of Discomfort Glare Across LED and HPS Light Sources The goal of this analysis was to determine whether differences in the perception of discomfort glare occurred between LED light sources of different CCTs, a legacy HPS light source, and unlighted conditions. One LMM was used to assess the effects of the type of light source on the perception of discomfort glare. For all statistical tests, age was included as a blocking factor. The level of significance was established at p < 0.05 for all statistical tests. Where relevant, post hoc analyses (pairwise com- parisons) were performed using Tukey’s honest significant difference. Least-squares means and standard errors were reported for LMMs. For MMLRs, odds ratios along with 95% confidence intervals were reported for significant factors. Photometric Analyses A photometric analysis was conducted with a calibrated photometer (ProMetric PM-9913E-1, Radiant Imaging®) to measure the changes in object luminance and contrast as the vehicles approached the object under different lighting conditions. The results of this analysis will help explain the influence of photometric properties of each lighting condition on a driver’s visual performance. Luminance and contrast measurements were conducted with the photometer mounted inside the experimental vehicle at a height of 5 ft (∼1.5 m) from the surface of the road (Figure 24). The Weber contrast formula was used to calculate the contrast of pedestrians and targets. Weber contrast is the difference between an object and its background relative to the luminance of the background and is calculated with the following formula: C L L L t b b ( )= − where C = luminance (Weber) contrast, Lt = object luminance, and Lb = background luminance. An object is considered be in negative contrast when it is darker than its background and posi- tive contrast when it is brighter than its background. For calculating the luminance of targets and Figure 24. Location of the calibrated photometer inside the test vehicle at the driver’s seat.

Research and Testing Results 51 pedestrians, polygons were traced around the pedestrian (Figure 25) and target, and the software calculated the mean luminance within the selected polygon. Background luminance was calcu- lated by tracing the same size polygon around the object location close to its boundaries (on the left, right, top, and bottom of the object of interest). The Weber contrast formula was used to calculate the contrast of pedestrians and targets. For pedestrians, the luminance and contrast were measured at distances of 820 ft (250 m), 410 ft (125 m), and 164 ft (50 m) from the pedestrian’s location. These distances constituted the 90th, 50th, and the 10th percentiles of the detection distances recorded for the pedestrians. Only the luminance and contrast of black- and gray-clothed pedestrians were measured, because most of the significant differences were between those two colors of clothing. Similarly, for targets, the luminance and contrast were measured at distances of 49 ft (15 m), 50 ft (164 m), and 328 ft (100 m), which constituted the 10th, 50th, and 90th percentiles of detection distances recorded during the visual performance analysis. In addition, the measured luminance and contrast val- ues of the pedestrians and targets were interpolated to determine the luminance and contrast at the instance of detection. Results Visual Performance Across LED Light Sources Pedestrian Detection Task—Detection Distance Analysis. All significant LMM results are summarized in Table 11. For ease of understanding, only the interactions in bold text are dis- cussed in the subsequent subsections. These interactions were selected because analyzing them can help in understanding the effects of all the relevant independent variables (type of light source, light level, surround ratio, and uniformity ratio) on the visual performance of drivers. Figure 25. Sample image output from the photometer and the method used to measure the luminance of the pedestrian by tracing polygons.

52 Solid-State Roadway Lighting Design Degrees of Freedom Source Numerator Denominator F-Value p-Value Age 1 33.3 30.2 <.0001 Surround ratio 1 34.2 18.8 0.000 Light level 2 2,082.5 26.2 <.0001 Speed 1 2,114.1 71.1 <.0001 Clothing color 3 2,082.2 61.7 <.0001 Offset 1 2,082.2 369.0 <.0001 Age surround ratio 1 34.3 5.8 0.022 Age speed 1 2,117.1 15.1 0.000 Age offset 1 2,082.2 29.9 <.0001 Surround ratio light level 2 2,082.5 4.2 0.015 Surround ratio clothing color 3 2,082.2 6.5 0.000 Type of light source speed 2 2,106.5 12.6 <.0001 Light level clothing color 6 2,082.2 14.8 <.0001 Speed clothing color 3 2,082.2 12.2 <.0001 Speed offset 1 2,082.2 15.7 <.0001 Clothing color offset 3 2,082.2 29.6 <.0001 Age surround ratio light level 2 2,082.5 6.8 0.001 Age surround ratio speed 1 2,117.9 15.7 <.0001 Age speed * clothing color 3 2,082.2 2.9 0.034 Uniformity ratio surround ratio speed 1 2,114.5 10.5 0.001 Surround ratio type of light source speed 2 2,106.8 3.6 0.028 Surround ratio type of light source offset 2 2,082.3 3.5 0.030 Surround ratio light level clothing color 6 2,082.2 9.1 <.0001 Surround ratio light level offset 2 2,082.3 9.7 <.0001 Surround ratio speed clothing color 3 2,082.2 8.7 <.0001 Light level speed offset 2 2,082.3 10.1 <.0001 Light level clothing color offset 6 2,082.3 14.9 <.0001 Speed clothing color offset 3 2,082.3 15.4 <.0001 Uniformity ratio surround ratio type of light source speed 2 2,106.6 3.2 0.041 Uniformity ratio surround ratio type of light source clothing color * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * age 6 2,082.3 2.2 0.044 Table 11. Significant statistical results from the LMM analysis of detection distance of LED light source comparison.

Research and Testing Results 53 Interactive Effect of Type of Light Source, Surround Ratio, and Offset. The combined effects of type of light source, surround ratio, and offset on detection distance are summarized in this section. There were no statistical differences between any of the LED light sources at either the offset location or under the surround ratio conditions, although the 4000K LED had slightly longer detection distances at the 2- and 10-ft offset locations under the high surround ratio condition and at the 10-ft offset location under the low surround ratio condition (Figure 26). Detection distance differences were statistically significant for the 4000K LED between the low and high surround ratio conditions at the 2-ft offset location. Further, detection distances were higher for the 2-ft offset location than for the 10-ft offset location for each light source and surround ratio condition (Figure 26). Pedestrian luminance increased as the vehicle approached the pedestrian locations under all surround ratios because of the vehicle headlamps (Figure 27). This increase in pedestrian luminance also altered the contrast (Figure 28) in which the pedestrian is rendered in because the luminance of the pedestrian increased (due to headlamps) without significantly influencing the background luminance. Pedestrian luminance (Figure 27) was greater under the higher surround ratio condition than under the lower surround ratio condition at both pedestrian locations. The pedestrian at the 2-ft offset location under the higher surround ratio was rendered in negative contrast until the vehicle was at distance of 300 ft from the pedestrian beyond which the pedestrian was rendered in positive contrast (Figure 28). When the pedestrians undergo a change in the contrast polarity, they go through a phase of contrast neutrality. Contrast neutrality occurs when the luminance of the pedestrian is equal to luminance of the background and contrast becomes zero. At zero contrast, the pedestrian becomes invisible to the driver. This change in the contrast polarity occurred only under the high surround ratio condition at approximately 300 ft from the pedes- trian at the 2-ft offset location. Under the higher surround ratio condition at the instant of the detection, the pedestrian was rendered in negative contrast, whereas under lower surround ratios, the pedestrian was always rendered in positive contrast (Figure 29). This negative contrast could have aided pedestrian detection under the higher surround ratio condition, because earlier research has shown that negatively contrasted objects are detected sooner than positively contrasted objects (Adrian 1989; Aulhorn 1964; Bhagavathula and Gibbons 2018). The pedestrian at the 10-ft offset loca- tion was always rendered in negative contrast throughout the vehicle’s approach to him or her (Figure 28). The differences in detection distance across the higher and lower surround ratios at Note: Values are means of detection distances and error bars indicate standard error LED 3000K LED 4000K LED 5000K LED 3000K LED 4000K LED 5000K 2 ft 10 ft Figure 26. Interactive effect of type of light source, surround ratio, and offset on detection distance.

54 Solid-State Roadway Lighting Design Pedestrian at 2 ft Right Distance (ft) Distance (ft) Pedestrian at 10 ft Right Legend HSR LSR Figure 27. Change in luminance as the vehicle approaches the pedestrian location. Pedestrian at 2 ft Right Distance (ft) Distance (ft) Pedestrian at 10 ft Right Legend HSR LSR Figure 28. Change in contrast as the vehicle approaches the pedestrian location.

Research and Testing Results 55 the 10-ft offset location were not as high as the differences at the 2-ft location, perhaps because the pedestrian at the 10-ft offset location was rendered in negative contrast under both surround ratio conditions (Figure 29). Although the pedestrian was rendered in negative contrast with the same magnitude, the luminance on the pedestrian at the higher surround ratio condition was higher than that at the lower surround ratio condition (Figure 27). This higher luminance under the higher surround ratio at the 10-ft offset location could have aided detection, thereby resulting in a longer detection distance. Interactive Effect of Type of Light Source, Surround Ratio, and Speed. The combined effects of type of light source, surround ratio, and speed on detection distance are summa- rized in this section. There were no statistically significant differences between LED light sources across the surround ratios and speeds. However, 4000K LED had the highest detection distance under the high surround ratio condition at 55 mph, and both 3000K and 4000K LED had the highest detection distance under the high surround ratio at 35 mph (Figure 30). Detection distance differences were statistically significant for the 4000K LED between the low and high surround ratio conditions at 55 mph. Detection distances were longer at 35 mph than at 55 mph for 3000K LED under low and high surround ratio conditions and for 5000K LED only at the high surround ratio condition. For 4000K LED, there were no differences in the detection distances between the 35 and 55 mph speeds. 2 ft right 2 ft right 10 ft right 10 ft right Figure 29. Average contrast of pedestrians at the instance of detection for each pedestrian location under both surround ratio conditions. 35 mph (56 km/h) 55 mph (89 km/h) LED 3000K LED 4000K LED 5000K LED 3000K LED 4000K LED 5000K Notes: Values are means of detection distances, and error bars indicate standard error. Figure 30. Interactive effect of type of light source, surround ratio, and speed on detection distance.

56 Solid-State Roadway Lighting Design Interactive Effect of Light Level, Surround Ratio, and Clothing Color. The combined effects of light level, surround ratio, and clothing color on detection distance are summarized in this section. In general, increases in the light level resulted in an increase in the detection distance, regardless of the surround ratio and the color of the clothing (Figure 31). Further, detection distances under the higher surround ratio were statistically greater than those under the lower surround ratio at each light level, especially for the black-colored clothing (Figure 31). Interactive Effect of Light Level, Surround Ratio, and Offset. The combined effects of light level, surround ratio, and offset on detection distance are summarized in this section. Increases in light level resulted in an increase in the detection distance under both of the surround ratios at the 2-ft offset location and under the high surround ratio at the 10-ft offset location (Figure 32). For the 2-ft offset location, detection distances under the high surround ratio were longer than those under the low surround ratio. For the 10-ft offset distance, the difference between the high surround ratio and low surround ratio was only statistically significant at the highest light level. Once again, at all light levels and surround ratios, detection distances at the 2-ft offset location were higher than those at the 10-ft offset location (Figure 32). Increases in the light level resulted in increases in the pedestrian luminance at both pedestrian offset locations (Figure 33). Higher surround ratio conditions also had higher pedestrian lumi- nance than lower surround ratio conditions. Pedestrians under the high surround ratio condi- tion for the 2-ft offset location were initially rendered in negative contrast, and as the vehicle got closer to the pedestrian, the contrast polarity reversed, and they were rendered in positive contrast (Figure 34). For the same offset location under a lower surround ratio, the pedestrian was always rendered in positive contrast. For the pedestrian at the 2-ft offset under the higher surround ratio, the light level affected the distance at which contrast neutrality (the phase of zero Note: Values are means of detection distances and error bars indicate standard error. Figure 31. Interactive effect of light level, surround ratio, and clothing color on detection distance.

Research and Testing Results 57 contrast when the negative contrast transitions to positive contrast) occurred (Figure 34). Con- trast neutrality occurred closest to the pedestrian location at the lowest light level and farthest from the pedestrian location at the medium light level. If pedestrians were rendered in negative contrast, this could have aided in their detection at the 2-ft offset location under the higher surround ratio (Figure 35). At the 10-ft offset location, pedestrians were rendered in almost similar negative contrast under all surround ratios (Fig- ure 35), but luminance was higher under the higher surround ratio condition, which could have contributed to the increase in detection distance. Detection distances in the 10-ft offset location were also lower than detection distances in the 2-ft offset location because they needed higher Note: Values are means of detection distances and error bars indicate standard error. Figure 32. Interactive effect of light level, surround ratio, and offset on detection distance. Pedestrian at 2 ft Right Distance (ft) Distance (ft) Distance (ft) Distance (ft) Pedestrian at 10 ft Right Light LevelHigh Low Medium Figure 33. Change in pedestrian luminance at different light levels as vehicles approached the pedestrian.

58 Solid-State Roadway Lighting Design contrast (more negative contrast) to be detected, as they were offset from the line of sight of the driver. This is consistent with existing research that showed that an increase in the offset distance from the roadway increases the threshold contrast for detection (Gibbons 1993). Interactive Effect of Uniformity Ratio, Surround Ratio, and Speed. The combined effects of uniformity ratio, surround ratio, and speed on detection distance are summarized in this section. There were no significant differences between low and high uniformity ratios across either the surround ratios or the speed limits. In general, detection distances were longer under Pedestrian at 2 ft Right Distance (ft) Distance (ft) Distance (ft) Distance (ft) Pedestrian at 10 ft Right Light Level High Low Medium Figure 34. Change in pedestrian contrast at different light levels as vehicles approached the pedestrian. HighMed. Low High Med. Low High Med.Low High Med.Low Figure 35. Average contrast of pedestrians at detection for each pedestrian location under both of the surround ratio conditions at the three light levels.

Research and Testing Results 59 the high surround ratio than under the low surround ratio (Figure 36). Detection distances were also longer in 35 mph conditions [mean = 505.21 ft (153.99 m)] than 55 mph conditions [mean = 434.28 ft (132.37 m)]. Interactive Effect of Age and Surround Ratio. The differences in detection distance between older and younger drivers were also dependent on the surround ratio. The detection distance between younger and older drivers was statistically significant only under the high surround ratio, with the younger drivers having longer detection distances than older drivers (Figure 37). There were no statistical differences between the older and younger drivers under the low sur- round ratio. Pedestrian Detection Task—Safe Stop Analysis. All significant MMLR results are summa- rized in Table 12. All the main effects except the effect of the type of light source were significant. The odds ratios of each of the significant factors and the 95% confidence intervals are shown in Table 13. For light level, the odds of detecting the pedestrian from a safe distance under the high light level were 1.7 and 1.4 times that of detecting the pedestrian from safe distance under low and medium light levels, respectively. This shows that the high level increases the odds of 35 mph (56 km/h) 55 mph (89 km/h) Note: Values are means of detection distances and error bars indicate standard error. Figure 36. Interactive effect of uniformity ratio, surround ratio, and speed on detection distance. Note: Values are means of detection distances and error bars indicate standard error. 0 100 200 300 400 500 600 700 800 HSR LSR Fe et Surround Ratio Old Young Figure 37. Interactive effect of age and surround ratio on detection distance.

60 Solid-State Roadway Lighting Design detecting a pedestrian. Higher surround ratios and higher uniformity ratios also increase the odds of detecting the pedestrian safely by 2.6 and 2.3 times, respectively. Younger driver were 5.6 times more likely to detect the pedestrian from a safe distance than older drivers. Regarding the color of clothing, pedestrians wearing red were three times more likely to be detected from a safe distance than those wearing black. Similarly, pedestrians wearing blue and gray were 1.3 and 1.6 times, respectively, more likely to be detected from a safe distance than those wearing black. Pedestrians that were located at the 2-ft offset were three times more likely to be detected from a safe distance than those who were at the 10-ft offset location. Finally, pedestrians were 37 times more likely to be detected from a safe distance when the drivers were traveling at 35 mph than at 55 mph. Target Detection Task. All significant LMM results are summarized in Table 14. For ease of understanding, only the interactions in bold text are discussed in the subsequent subsections. Interactive Effect of Type of Light Source, Light Level, and Uniformity Ratio. The com- bined effects of the type of light source, light level, and uniformity ratio on detection distance are summarized in this section. Uniformity ratios across LED light sources and light levels were not Effect Numerator Denominator F-Value p-Value Degrees of Freedom Light level 2 2,383 6.8 Surround ratio 1 2,383 8.0 Uniformity ratio 1 2,383 6.2 Age 1 2,383 22.5 Clothing color 3 2,383 17.7 Offset 1 2,383 99.5 Speed 1 2,383 546.8 0.001 0.005 0.013 <.0001 <.0001 <.0001 <.0001 Table 12. Significant statistical results from the MMLR analysis of type of stop of pedestrians under LED light source comparison. Independent Variable Level Reference Level Odds Ratio 95% Confidence Limits Light level High Low 1.7 1.3–2.3 High Medium 1.4 1.1–1.8 Surround ratio HSR LSR 2.6 1.3–5.0 Uniformity ratio HUR LUR 2.3 1.2–4.4 Age Younger Older 5.6 2.7–11.4 Color Blue Black 1.3 1.0–1.8 Gray Black 1.6 1.2–2.2 Red Black 3.3 2.4–4.6 Offset 2 ft 10 ft 3.4 2.7–4.3 Speed 35 mph 55 mph 37.7 27.8–51.1 Table 13. Odds ratios of significant factors in MMLR analysis of detection of pedestrians.

Research and Testing Results 61 * Degrees of Freedom Source Numerator Denominator F-Value p-Value Age 1 33 34.6 Light level 2 1,368.1 46.2 Speed 1 1,395 26.6 Target color 2 1,367.5 13.5 Offset 1 1,368.1 19.0 Age light level 2 1,367.9 5.9 Age target color 2 1,367.5 4.7 Age offset 1 1,367.7 7.2 Uniformity ratio target color 2 1,367.2 10.0 Uniformity ratio offset 1 1,367.9 16.4 Surround ratio target color 2 1,367.4 10.4 Type of light source light level 4 1,367.9 2.5 Type of light source speed 2 1,385.3 11.5 Light level speed 2 1,367.6 16.2 Light level target color 4 1,367.4 29.9 Light level offset 2 1,367.5 7.4 Clothing color offset 2 1,367.2 12.3 Age uniformity ratio target color 2 1,367.2 3.2 Age uniformity ratio offset 1 1,367.5 13.2 Age type of light source speed 2 1,397.8 3.7 Age speed target color 2 1,367.3 3.5 Uniformity ratio surround ratio offset 1 1,367.1 7.3 Uniformity ratio type of light source light level 4 1,367.6 3.0 Uniformity ratio type of light source target color 4 1,367.1 2.5 Uniformity ratio type of light source offset 2 1,367.3 4.5 Uniformity ratio target color 2 1,366.9 5.3 Surround ratio type of light source light level 4 1,367.5 2.7 Type of light source target color offset 4 1,367 3.7 Light level speed target color 4 1,367.6 35.6 Light level speed offset 2 1,367.3 4.6 Light level target color offset 4 1,367.3 20.1 Speed target color * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * offset * * * * * * * * offset 2 1,367.1 12.1 <.0001 <.0001 <.0001 <.0001 <.0001 0.003 0.009 0.007 <.0001 <.0001 <.0001 0.039 <.0001 <.0001 <.0001 0.001 <.0001 0.041 0.000 0.026 0.030 0.007 0.019 0.039 0.012 0.005 0.028 0.005 <.0001 0.010 <.0001 <.0001 Table 14. Significant statistical results from the LMM analysis of detection distance of targets under LED light source comparison.

62 Solid-State Roadway Lighting Design different. An increase in light levels across all types of light sources and uniformity ratios resulted in longer detection distances (Figure 38). There were also no statistical differences between light types across the uniformity ratios and light levels. Interactive Effect of Type of Light Source, Light Level, and Surround Ratio. There were no statistical differences between the surround ratios across the LED light sources and light levels. There were also no statistical differences between LED light sources across the surround ratios and light levels. At medium and high light levels under a high surround ratio, slightly higher detection distances were found for 4000K LED (Figure 39). Increases in light levels resulted in an increase in detection distances across all the light types and surround ratios. The luminance and the contrasts of the targets in both surround ratios changed as the vehicle approached the target locations because of increased illumination from the vehicle headlamps. Target luminance values were slightly higher under the higher surround ratio as compared with the lower surround ratio (Figure 40). The targets were initially rendered in negative contrast and transitioned to positive contrast as the vehicle approached the target location (Figure 41). The distance at which the change in the contrast polarity occurred—that is, the point at which the target passed through a phase of contrast neutrality—depended on the prevailing light level. The higher the light level, the closer the target was to the vehicle when contrast neutrality occurred (Figure 41). The lack of significant differences between the detection distances at each surround ratio could be explained by the absence of major differences between the contrasts of the targets at detection (Figure 42). At each light level, the contrasts between the surround ratios were of the same polarity. However, as the light level increased, the targets were detected from farther away LED 3000K LED 4000K LED 5000K LED 3000K LED 4000K LED 5000K LED 3000K LED 4000K LED 5000K Note: Values are means of detection distances and error bars indicate standard error. Figure 38. Interactive effect of type of light source, light level, and uniformity ratio on detection distance.

Research and Testing Results 63 LED 3000K LED 4000K LED 5000K LED 3000K LED 4000K LED 5000K Note: Values are means of detection distances and error bars represent standard errors. Figure 39. Interactive effect of light type, light level, and surround ratio on detection distance of targets. Light Level High Low Medium Figure 40. Change in target luminance as the vehicle approaches the target under different light levels. and, as a result, the contrast at which they were detected also differed widely in magnitude and polarity (Figure 42). Targets under the highest light level were detected from the farthest distance when they were rendered in negative contrast (Figure 42). At the medium light level, the targets were still rendered in negative contrast but were much lower in magnitude than under the high- est light level and, as a result, were detected from shorter distances. Under the lowest light level, the targets were detected when they were rendered in positive contrast at the shortest detection distance (Figure 42). Interactive Effect of Type of Light Source, Uniformity Ratio, and Target Color. There were no significant differences between LED light sources across uniformity ratios and target colors. There were also no significant differences between uniformity ratios across all the LED light sources and target colors. At the higher uniformity ratio for 4000K LED, detection distances for

64 Solid-State Roadway Lighting Design blue-colored targets were longer than those for gray- and red-colored targets (Figure 43). Simi- larly, for the 3000K LED, detection distances for blue-colored targets were longer than those for gray-colored targets under the high uniformity ratio. The luminance and the contrast of targets of all colors changed as the vehicle approached their location (Figure 44). Blue and red targets were initially rendered in negative contrast and transitioned to positive contrast as the vehicle approached their location. The gray target was always rendered in positive contrast. On average, at detection, the blue and red targets were rendered in negative contrast, and the gray target was rendered in positive contrast (Figure 45). The rendering of the blue target Light Level High Low Medium Figure 41. Change in target contrast as the vehicle approaches the target under different light levels. Figure 42. Average target contrast at detection at each light level.

Research and Testing Results 65 0 50 100 150 200 250 300 Blue Gray Red D et ec tio n D is ta nc e (f t) Target Color LED 3000K LED 4000K LED 5000K Note: Values are means of detection distances and error bars represent standard error. Figure 43. Interactive effects of type of light source and target color at the high uniformity ratio. Color Blue Gray Red Figure 44. Change in target luminance and contrast across the three colored targets as the vehicle approached the target.

66 Solid-State Roadway Lighting Design in higher negative contrast could have resulted in a longer detection distance than that of the gray target, which was rendered in positive contrast. Further, the contrast of the blue target was higher in magnitude than that of the red target (Figure 45) at detection, which could have resulted in the blue target having a longer detection distance than the red target. Interactive Effect of Light Level, Speed, and Offset. Significant differences in detection distance were observed at the lowest light level across both offset distances. The detection distance at 35 mph was longer than that at 55 mph (Figure 46). At the medium light level, these differences in detection distance across lower and higher speeds were observed only for the targets on the right-hand side (Figure 46); at the highest light level, there were no differences between detection distances across either speed (Figure 46). Interactive Effect of Age and Light Level. At all light levels, detection distances were sig- nificantly longer for younger drivers than for older drivers (Figure 47). For both age groups, the differences between high and low and between high and medium light levels were statistically significant (Figure 47). Target Detection Task—Safe Stop Analysis. All significant MMLR results are summarized in Table 15. The main effects of light level, age, offset, and speed were significant. Figure 45. Average contrast of the blue-, gray-, and red-colored targets at detection. Note: Values are means of detection distances and error bars represent standard error. Figure 46. Interactive effect of light level, speed, and offset on detection distance of targets.

Research and Testing Results 67 The odds ratio estimates of each of the significant factors and the 95% confidence intervals are shown in Table 16. As with the pedestrian detection task, the odds of detecting the targets from a safe distance were 1.6 times higher under the high light level as compared with the medium and low light levels. Younger drivers were almost four times more likely to detect the target from a safe distance than older drivers. Targets on the left were 1.4 times more likely to be detected from a safe distance than those on the right. Targets were almost 20 times more likely to be detected from a safe distance when drivers were traveling at 35 mph rather than at 55 mph. Visual Performance Across LED, HPS, and Unlighted Conditions Pedestrian Detection Task—Detection Distance Analysis. All significant LMM results are summarized in Table 17. The main effects of age, type of light source, and offset were significant. There were no significant differences between any of the light sources. The only significant differences were those between each light source and the unlighted con- dition. The detection distance under the unlighted condition was the lowest as compared with all of the light sources (Figure 48). Older drivers had shorter detection distances [M = 431.46 ft (131.51 m)] than younger drivers [M = 579.10 ft (176.51 m)], and pedestrians at a 2-ft offset distance had longer detection distances [M = 567.65 ft (173.02 m)] than those at a 10-ft offset [M = 442.91 ft (135.00 m)]. 0 50 100 150 200 250 300 Older Younger Age Low Medium High D et ec tio n D is ta nc e (f t) Note: Values are means of detection distances and error bars represent standard error. Figure 47. Interactive effect of age and light level on detection distance of targets. Degrees of Freedom Effect Numerator Denominator F-Value p-Value Light level 2 1,742 3.9 0.021 Age 1 1,742 19.5 <.0001 Offset 1 1,742 4.7 0.031 Speed 1 1,742 148.8 <.0001 Table 15. Significant statistical results from the MMLR analysis of type of stop of targets under LED light source comparison.

68 Solid-State Roadway Lighting Design Independent Variable Level Reference Level Odds Ratio 95% Confidence Limits Light level High Medium 1.6 1.1–2.3 High Low 1.6 1.1–2.3 Age Younger Older 3.8 2.1–6.8 Offset 2 ft left 2 ft right 1.4 1.0–1.9 Speed 35 mph 55 mph 20.2 12.5–32.8 Table 16. Odds ratios of significant factors in MMLR analysis of detection of pedestrians. Degrees of Freedom Source Numerator Denominator F-Value p-Value Age 1 18.2 14.0 0.002 Type of light source 4 69.7 48.4 <.0001 Offset 1 351.2 39.2 <.0001 Table 17. Significant statistical results from the LMM analysis of detection distance of pedestrians under LED and HPS light source comparison. LED 3000K LED 4000K LED 5000K Type of Light Source HPS Unlit D et ec tio n D is ta nc e (f t) Note: Values are means of detection distances and error bars represent standard error. Figure 48. Effect of type of light source on the detection distance of pedestrians.

Research and Testing Results 69 Pedestrian Detection Task—Safe Stop Analysis. All significant MMLR results are summa- rized in Table 18. All the main effects were significant. The odds ratio estimates of each of the significant factors and the 95% confidence intervals are shown in Table 19. Detection of pedestrians at a safe stopping distance were higher under the lighted conditions than under the unlighted conditions. There were no statistical differences between the types of light sources used in the study in terms of detection of pedestrians from a safe distance. Younger drivers were 10 times more likely to detect the pedestrian from a safe distance than older drivers. Pedestrians in red-colored clothing were almost three times as likely to be detected from a safe distance as those in black-colored clothing. Interestingly, the odds of detecting a pedestrian in blue clothing were lower than the odds of detecting a pedestrian in black clothing. Pedestrians at an offset of 2 ft were two times more likely to be detected from a safe distance than those at a 10-ft offset. Drivers traveling at the 35 mph speed limit were 94 times more likely to detect a pedestrian from a safe distance than those traveling at 55 mph. Target Detection Task. All significant LMM results are summarized in Table 20. Only the main effects of age and type of light source were significant. Degrees of Freedom Source Numerator Denominator F-Value p-Value Type of light source 4 396 17.2 <.0001 Age 1 396 9.0 0.003 Clothing color 3 396 3.5 0.015 Offset 1 396 5.4 0.021 Speed 1 396 55.2 <.0001 Table 18. Significant statistical results from the MMLR analysis of detection of pedestrians at a safe stopping distance under LED and HPS light sources. Independent Variable Level Reference Level Odds Ratio 95% Confidence Limits Type of light source 3000K LED Unlit 45.1 11.2–181.4 4000K LED Unlit 259.6 55.6 to >999.9 5000K LED Unlit 225.9 27.0 to >999.9 2100K HPS Unlit >999.9 143.9 to >999.9 Age Younger Older 9.9 2.2–44.5 Color Blue Black 0.7 0.3–1.5 Gray Black 1.0 0.4–2.5 Red Black 2.9 1.2–7.4 Offset 2 ft 10 ft 2.1 101–4.1 Speed 35 mph 55 mph 94.5 28.4–314.9 Table 19. Odds ratios of significant factors in MMLR analysis of detection of pedestrians.

70 Solid-State Roadway Lighting Design Of all the light sources, 3000K LED had the lowest detection distance, and the differences between 3000K and 4000K LED were statistically significant (Figure 49). The difference in detec- tion distance between 4000K LED and unlighted conditions was also statistically significant. Interestingly, the differences between 4000K LED, 5000K LED, and HPS light sources were not statistically significant, even though HPS had the highest detection distance (Figure 49). The difference in detection distance between HPS and unlighted conditions also was not statistically significant. Target Detection Task—Type of Stop Analysis. The MMLR model for the type of stop for targets did not converge. Visual Performance Across Uniformity Ratios of LED and HPS Light Sources. For both pedestrian and target detection LMMs, the effect of the uniformity ratio on detection distance was not significant. Perception of Discomfort Glare Across LED Light Sources. All significant LMM results are summarized in Table 21. Only the main effects of light level and a two-way interaction involving age and type of light source were significant. For the two interactions involving age and type of light source, there were no statistical dif- ferences observed between older and younger drivers across the LED light sources. There were also no differences in the glare ratings across each type of light for each age group. Increasing the light level resulted in an increase in the glare rating, with the only significant difference being Degrees of Freedom Source Numerator Denominator F-Value p-Value Age 1 15.2 11.3 0.004 Type of light source 4 67.8 4.5 0.003 Table 20. Significant statistical results from the LMM analysis of detection distance of targets under LED and HPS light source. LED 3000K LED 4000K LED 5000K Type of Light Source HPS Unlit D et ec tio n D is ta nc e (f t) Note: Values are means of detection distances and error bars represent standard errors. Figure 49. Effect of type of light source on the detection distance of targets.

Research and Testing Results 71 between high and low light levels, the high light level having a higher glare rating than the low light level (Figure 50). Perception of Discomfort Glare Across LED and HPS Light Sources. The LMM analysis revealed that none of the factors were significant, which suggests no differences between LED light sources and the HPS light source. Glare ratings were not measured in the unlighted condition. Discussion The goals of this study were to determine whether a drivers’ visual performance (measured by using pedestrian and target detection tasks) and perception of discomfort glare (measured by a rating scale) were affected by different types of light sources (3000K, 4000K, 5000K LEDs, and HPS), light levels, surround ratios, and uniformity ratios. Five major findings were evident: 1. There were no major statistical differences in visual performance across any of the light sources. 2. Increasing the light level increased visual performance. 3. Higher surround ratios increased drivers’ visual performance. 4. Uniformity ratios did not seem to influence drivers’ visual performance. 5. None of the light sources was a major source of discomfort glare for drivers. Effect of Type of Light In both the pedestrian and target detection tasks, across all the light sources used in the study, there were no statistical differences between the detection distances of drivers or drivers’ ability Degrees of Freedom Source Numerator Denominator F-Value p-Value Light level 2 203.2 6.0 0.003 Age * type of light source 2 38.3 4.8 0.014 Table 21. Significant statistical results from the LMM analysis of perception of discomfort glare under LED light sources. Figure 50. Effect of light level on discomfort glare rating.

72 Solid-State Roadway Lighting Design to detect the pedestrians or targets from a safe distance. The only statistically significant dif- ferences in the visual performance of pedestrian and target detection were those between the lighted and unlighted conditions. There was, however, a practical difference for 4000K, for which detection distances were greater, especially for pedestrians, than any of the other light sources. The detection distances were statistically longer under every light source when com- pared with the unlighted condition. Similarly, the odds of detection from a safe distance were higher under every light source than under the unlighted condition. There were some dif- ferences in the detection of targets: detection distances of targets under 3000K LED were the lowest and were not statistically different from the detection distances under the unlighted condition. Overall, for pedestrian detection tasks, detection distances under 4000K LED were slightly higher, albeit not statistically significant, especially under the higher surround ratio and higher speed conditions. These results could indicate that the 4000K LED light could increase the visual performance of drivers at higher speeds. At the medium light level, HPS light sources had higher detection distances, but this was not statistically different from the detection dis- tances under LED light sources such as 3000K, 4000K, and 5000K LED. The results indicate that the SPD of the light source does not significantly influence driver visual performance at speeds greater than 35 mph. Effect of Light Level Increases in the light level increased visual performance, as evidenced by the increase in detection distances. Increases in the light level also increased the odds of detection from a safe distance. This result is consistent with earlier research on the effects of illuminance on visual performance (Bhagavathula and Gibbons 2015; Bhagavathula et al. 2018; Fotios and Cheal 2009; Gibbons et al. 2016; Gibbons et al. 2008). The increase in visual performance with the increase in light level was observed across both detection tasks. Increase in the visual perfor- mance with the increase of light level (especially for detecting the pedestrians at the 2-ft offset location and all target locations) was likely because the change increased the negative contrast in which both the pedestrians and targets were rendered. Further, the increase in the light level increased the luminance on the pedestrians and targets and affected the distance at which the contrast neutrality occurred. It is important for the contrast polarity to occur farthest from the location of the object or not to occur at all because, during contrast polarity, the object goes through contrast neutrality (i.e., the contrast is zero because the target luminance is equal to the background luminance), at which point the object becomes invisible to the driver. If this transition occurs closer to the object, it gives the driver less time to react than when it occurs farther from the object. The increase in light level increases the amount of visual information a driver can gather from the environment and can result in better visual performance. Most of the significant differences between the light levels were between high and low and high and medium. There were no differences between medium and low light levels; this could be because of differ- ences in average luminance across medium and low light levels. The difference in luminance between medium and low light levels was only 0.3 cd/m2 and was likely not enough to produce a statistical difference in visual performance. The differences between average luminance for the high to low and high and medium light level comparisons were 0.8 and 0.5 cd/m2, respec- tively, which could have produced statistically significant results. The increase in light level also increased the detection distances at 55 mph and the low surround ratios, especially for the pedestrians. These results indicate that, on higher-speed roadways, higher light levels aid driver visual performance and could potentially aid pedestrian safety. While pedestrians are not typical on higher-speed roadways, it should be noted that the same benefit is expected for an animal crossing.

Research and Testing Results 73 Effect of Surround Ratio Surround ratios significantly affected driver visual performance. Increases in the surround ratio increased visual performance, as evidenced by longer detection distances and increased odds of detection from a safe distance. The effect of surround ratio was predominantly observed in the pedestrian detection task because it was the only task that had the furthermost offset pre- sentation (10 ft). The increase in detection under higher surround ratios was observed across all LED light sources but was more pronounced under the 4000K LED for the pedestrian detec- tion task at an offset distance of 10 ft and at the higher speed (55 mph). Higher surround ratios increase the amount of light incident outside the roadway, and this additional illumination could help drivers gather more visual information from areas adjacent to the roadway. Also, at higher speeds, the cone of drivers’ field of vision narrows (FHWA 2015), and additional illumination under the high surround ratio seems to benefit the driver based on the results of this study. High visual performance under the higher surround ratios (especially at the 2-ft offset loca- tion) could be attributed to both pedestrians and targets being rendered in negative contrast. Existing research (Adrian 1989; Aulhorn 1964; Bhagavathula et al. 2018) shows that objects rendered in negative contrast were detected sooner, or from farther away, than those rendered in positive contrast. Effect of Uniformity Ratio Uniformity ratio did not seem to influence driver detection distance in either the pedestrian or target detection tasks. However, for the pedestrian detection task in the type of stop analy- sis, results showed that drivers were more likely to detect the pedestrian from a safe distance under higher uniformity ratios than under lower uniformity ratios. These results indicate that higher uniformity ratios (less uniform roadways) aid the detection of pedestrians at night. It should be noted that the uniformity ratios studied in the project were relatively similar and within the AASHTO recommendations. A larger uniformity gap might yield different results. The lack of difference in the uniformity ratios in the detection distance analysis could be because uniformity of the road affects drivers’ visual comfort more than their visual perfor- mance. These results also indicate that further research is required to understand the effects of uniformity on visual performance. Perception of Discomfort Glare There were no significant effects of type of light source, surround ratio, and uniformity ratio on the perception of discomfort glare. Only light level had a significant effect on the perception of discomfort glare: the increase in the light level increased the discomfort glare rating. However, it is important to note that, even at the highest light level, the glare rating was less than “noticeable” (i.e., less than 2 on a scale ranging from 0 to 6, where 6 is “intolerable”). Effects of Offset, Speed, Age, and Color An increase in offset distance and speed lowered drivers’ visual performance, as observed by lower detection distances and decreasing odds of detection from a safe distance. This decrease in visual performance could be attributed to the reduction in the drivers’ field of view with the increase in speed (FHWA 2015). The visual performance of younger drivers was better than that of older drivers, as evidenced by longer detection distances and higher odds of detection from a safe distance. This result is consistent with existing research in this area (Bhagavathula et al. 2018). The decrease in visual performance among older drivers was likely due to age-related physiological changes in the eyes that lead to reduced visual acuity and contrast sensitivity. The lowest detection distance was for pedestrians in black-colored clothing, who also had the

74 Solid-State Roadway Lighting Design lowest odds of being detected from a safe distance. These results also indicate that pedestrians should refrain from wearing black-colored clothing in nighttime roadway environments to increase their visibility. Effects of Headlamps on Object Luminance and Contrast Pedestrians and targets underwent changes in luminance and contrast as the vehicles approached their locations. The increase in luminance was dependent on the prevailing light level and surround and uniformity ratio conditions. In general, target luminance increased as the vehicle got closer because targets entered the effective range of the vehicle headlamps. Light levels and surround ratios also greatly influenced the magnitude and polarity of contrast in which pedestrians and targets were rendered. All target locations under all surround ratios and pedestrians at the 2-ft offset locations under high surround ratios underwent changes in contrast polarity as the vehicle approached their location. During these transitions in contrast polarity, objects became temporarily invisible as the luminance of the object became equal to the lumi- nance of the background. It is important to minimize these transitions so that drivers can always locate hazards on the roadway. Conclusion The results of the Smart Road testing can be summarized as follows: 1. The surround ratio plays an important role in augmenting the visual performance of the driver. Increasing the surround ratio by illuminating the areas adjacent to the roadway increased drivers’ visual performance. Increasing the surround ratio also increased the visual performance of drivers of all age groups at both tested speeds. 2. Increasing the light level also increases drivers’ visual performance, particularly at higher speeds. Higher light levels also increased driver visual performance under the lower surround ratio conditions. 3. The luminance of an object increases as a vehicle approaches it. The photometric analysis showed that target and pedestrian luminance increased as the vehicle approached. Target and pedestrian contrasts were also affected by vehicle headlamps and the location of the vehicle with respect to the location of the object. The location of some pedestrians and targets also changed the polarity of contrast in which they were rendered (i.e., starting in negative con- trast and then transitioning to positive contrast as the vehicle approached the pedestrian or target location). A good roadway lighting design should ensure that this transition happens farther from the pedestrian/target location or does not happen at all, so that the driver can always see the hazard on the roadway. 4. Better visual performance under higher surround ratios and higher light levels is the result of the negative contrast in which the objects are rendered under those conditions. As shown by the photometric analysis, negative contrast resulted in better visual performance as compared with positive contrast. Increases in the magnitude of the negative contrast also resulted in better visual performance. 5. Under most conditions, the color temperature of an LED or HPS light source does not affect drivers’ visual performance. No major differences were observed between the correlated color temperatures of LED light sources or between the LED and HPS light sources in terms of driver visual performance under most conditions. However, drivers’ visual performance under 4000K LED was the highest at both tested speeds and at higher surround ratios. 6. An increase in the uniformity ratio of the roadway also increases the odds of detection from a safe distance. 7. None of the evaluated light sources (LEDs of all CCTs and HPS) is a significant source of discomfort glare. The ratings of discomfort glare were mainly affected by the light levels, and even at the highest light level evaluated, all were lower than “noticeable.”

Research and Testing Results 75 Research Roadmap The following areas of research should be considered to better understand the impacts and benefits of SSL. Areas of funding include NCHRP and FHWA. • Metrics for assessing the effectiveness of roadway lighting, with and without headlights, to achieve adequate contrast. Several factors are included in this research problem. One of those is what task is being used to assess the effectiveness of a lighting system. While the ultimate goal is to reduce crashes and fatalities on roadways, the primary task on a street may be seeing a pedestrian or cyclist, and the primary task on a highway may be remaining on the roadway and identifying geometric changes and off- and on-ramps. Contrast may be the proper met- ric for determining effectiveness, but other risk factors must be considered. IES has a task force(which includes members of this project team) that is examining this issue. The work of the task force and other work could help in addressing some of these items in future research. Approximate research cost: $200,000. • Standardization and asset management for SSL replacement to simplify maintenance. Asset management for roadway lighting systems is done in a number of ways. At times, a system assessment is performed, and data are collected for an entire lighting system and incor- porated into a geographic information system with complete descriptions of the asset. If an adaptive lighting system is installed, mapping can be done by installing the adaptive lighting system nodes on the luminaires to self-identify location, wattage, operational status, and con- trol scheme. Once this baseline information and mapping are done for the system, structural and electrical assessments can be made for the pole, arms, and distribution system. Strict standardization may be difficult from state to state, but recommendations can be made part of these guidelines and of future research into other areas. Approximate research cost: $150,000. • Potential revenue streams from data collection and barriers to their implementation. Sev- eral methods of providing revenue streams from street lighting assets have been investigated, including simply considering the street lighting system as real estate that can be leased for equipment like 4G wireless. Other methods look ahead to use of streetlight assets by vehicle- to-vehicle (V2V) and vehicle-to-infrastructure (V2I) systems as a network base for collecting and transmitting data, monitoring a variety of security or urban management systems, and integrating with Smart city components, including water monitoring and other city system management. These methods are, for the most part, concepts at this point but may become more viable as new technology is deployed. They also depend on the infrastructure in place, such as pole breakaway devices, switched or unswitched power, and pole designs. Approximate research cost: $50,000. • Real-time asset management within the context of the ITS framework. As with the potential revenue streams noted above, it is likely that the ITS framework will be integrated with street- light systems and control. In essence, the ITS can serve as an integrated system to – Maintain inventories of lighting systems and interacting asset classes (e.g., high-mast poles, signs) with geospatial locations, – Monitor key attributes and identify faulty conditions in real time for corrective actions, – Track maintenance histories and associated costs, and – Track life expectancies of lighting systems. In addition, the real-time data capture has the potential to supply other performance metrics, such as safety, mobility, and the performance of interacting asset classes and operating condi- tions (e.g., wet conditions). NCHRP Project 08-36, Task 114, “Transportation Asset Manage- ment for Ancillary Structures,” provides guidance on incorporating ancillary assets into an

76 Solid-State Roadway Lighting Design integrated asset management program and can serve as a starting point for exploring real-time management of lighting systems within the ITS framework. Future research will be needed to address integration with Smart city systems and connected and autonomous vehicle systems. Approximate research cost: $70,000. • Metrics to characterize the environment and health effects of the roadway light spectrum, amount, distribution, and timing. A fair amount of research is available on the issue of light at night and the environment and effect on people’s health. Unfortunately, opinions on the results of this research and how recommendations are crafted from it vary. A cohesive plan that directly correlates roadway lighting with safety needs and the impact of lighting on health and the environment would be valuable. The precise limits at which impacts can occur is likely to change as further research in this area is done. Approximate research cost: $350,000. • Metrics for assessing light depreciation and means of determining the end of useful life. Currently, LED life is based on the point in time at which the lumen output of the source depreciates by 30%. It does not consider the expected life of the equipment or of the drivers and controllers, which is much shorter. These components are often the true indicator of when a luminaire will stop operating. Some existing metrics include the use of MTBF data, which may be better suited to SSL. Research in this area to validate this and other metrics would be valuable. Approximate research cost: $300,000. • Crash data analysis of HID technologies versus SSL technologies. A review of historical crash data for various roadway types, interchanges, intersections, and roundabouts to deter- mine whether any changes in day-to-night crash rates occur with the use of SSL lighting is required. SSL luminaires have not been in the field long enough for statistical analyses to show any differences. Approximate research cost: $100,000.