Guidelines for Nighttime Visibility of Overhead Signs (2016)

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.

27 Open-Road Study With the understanding that sign lighting and increased luminance provide no marginal benefit to legibility when drivers are in conditions with low visual complexity, the research team investigated whether or not these factors affected how well nighttime drivers read signs on the open road in more visually complex settings. The closed-course study provided a controlled environment where the tested effects were inves- tigated in a full-factorial type of experiment. The open-road study was able to test drivers in more diverse conditions of visual complexity and workload but at the expense of the tight control and factorial design that was used in the closed-course study. The open-road study was completed in 2013. Experimental Design Data for the open-road study were collected in three different locations: Bryan/College Station, Texas; San Antonio, Texas; and Orlando, Florida. With an emphasis on different levels of environmental or background complexity and sign luminance, routes were selected in each location to contain a variety of overhead and shoulder-mounted guide and street name signs situated with various levels of complexity and luminance. The background complexity was affected by roadway alignment, adjacent signs, adjacent commercial and residential buildings, and varying light sources that affect the ambient light and can potentially cause glare. Vehicular traffic impacts background complexity and luminance, but this could not be controlled from one participant to another. To account for this variability as best as possible, the driving tests were only performed Sun- day through Thursday after evening rush-hour traffic under uncongested driving conditions. The study protocol was iden- tical between the cities with the exception of the route. Study Routes The three study routes in Bryan/College Station (B/CS), San Antonio, and Orlando were selected to take advantage of the variation in practices for sign lighting and sheeting materials and levels of background complexity and ambient lighting. The routes were selected so that each drive lasted between 30 and 60 min (see Appendix C for additional details). While the closed-course study focused exclusively on overhead guide signs, the open-road study had drivers read overhead and shoulder-mounted guide signs and street name signs. The inclusion of multiple sign types and placement locations resulted in a typical driving experience that encompasses route finding. Examples of the environments encountered on the open road are shown in Figure 19. The cities where data were collected employ different prac- tices to ensure the signs are visible. Orlando is the only city where the route included overhead guide signs that are lit. Each city has internally illuminated overhead street name signs in addition to unlit overhead street name signs. With different sheeting products used, diversity in lighting practices, and light provided by the surrounding environment in which each sign is located, the open-road study contained a vari- ety of features that influence visibility and how well drivers read a sign. The inability to control for specific factors such as sign lighting or sign sheeting resulted in requiring the use of measured sign luminance as an independent variable, as described in the Analysis section that follows. Procedure Seventy-three participants drove in the study, with an even split between male and female participants. There were twice as many older drivers (55+ years) than younger drivers (18– 35 years). The number of participants was split nearly even across the three locations. Upon meeting the research team and passing tests of visual acuity, contrast sensitivity, and color blindness, each participant entered the study vehicle, a 2006 Toyota Highlander, and adjusted basic driver settings. The vehicle was instrumented with a GPS receiver for tracking the position of the vehicle and cameras for logging photometric C H A P T E R 4

28 images. Each study participant drove the vehicle through the designated route. Routing guidance was provided by a researcher in the front passenger seat who also watched for potential traffic conflicts and recorded any relevant comments from the participant. A second researcher sat in the back seat to monitor the data collection equipment and indicate in the data stream when the participant correctly read a sign. Throughout the drive, the researcher in the passenger seat provided directions and requested that the participant indi- cate the moment when he or she could read a sign with the corresponding street. The participants were asked to respond when they were confident in their response. For uniformity in the experience from one driver to the next, the researcher instructed the participant to select a specific lane when multi- ple lanes were available. Appendix C contains the specific instructions provided to the research participants. Figure 20 shows images of the study vehicle and data collec- tion equipment used for the open-road study. Data The procedure for measuring sign luminance varied from that of the closed-course study because the signs were located on the open road and the vehicle containing the photom- eter could not stop throughout the procedure. During the open-road study, a vehicle-mounted photometer captured images continuously in quick succession while the vehi- cle approached each sign. The image from a distance to the (a) High Complexity (Overhead Guide Sign) (b) Low Complexity (Overhead Guide Sign) (c) High Complexity (Overhead Street Name Sign) (d) Low Complexity (Overhead Street Name Sign) Figure 19. Images of guide and street name signs from the study routes in environments of various background complexity.

29 sign most closely representing 40 ft/in. of letter height was extracted for measuring the luminance of the sign’s legend and background. The visual complexity of the scene approaching the study signs was of interest in the open-road study. The researchers were interested in studying how the complexity of the scenes would impact the distance at which drivers could read the signs. The scene complexity for each sign was determined with the image processing technique discussed in Appendix B. The same images used to measure the sign luminance were evalu- ated with image processing software that extracted elements of the scene such as the number of light sources, number of objects, and overall image saturation. A level of complexity on a 1–5 scale was assigned to each image based on statistical models validated with data collected from a survey of people who viewed the images and provided subjective complexity ratings. The purpose of the image processing software was to remove the element of subjectivity from the determination of rating visual complexity. Figure 21 shows values of legend luminance for the differ- ent samples. Luminance from both the open-road courses and the closed-course study are shown for comparisons of what the drivers experienced. There is a wide distribution of luminance values across the different locations where data were collected. Table 11 contains summary statistics that include the levels of complexity encountered. The visual complexity for shoulder- mounted guide signs was comparable to that of overhead guide signs, except in Orlando where the shoulder-mounted guide signs were in locations with greater complexity. The complexity tends to increase from Bryan/College Station, to San Antonio, to Orlando. Table 12 contains the luminance values of the signs as iden- tified in Figure 21. One important feature in Table 12 is the leg- end height. Because the legend height was not constant across all signs, the dependent variable in the analyses was the recog- nition index, measured in feet of recognition distance per inch of letter height. This reduced the need to include the legend (a) Instrumented 2006 Toyota Highlander (b) Instrumentation Setup (c) Camera, Internal V-Lambda Filter, and Lens (d) Camera System with GPS Figure 20. Open-road data collection equipment.

30 Figure 21. Legend luminance values of signs in the open- and closed-course studies. 0 10 20 30 40 50 Le ge nd L um in an ce (c d/ m 2 ) B/CS Orlando San Antonio Closed Course Unlit Overhead Guide Signs Shoulder Guide Signs Overhead Street Name Signs Lit Unlit Lit Shoulder Street Name Signs Visual Complexity Location Total Signs Sign Type (qty.) Min. Avg. Max. Bryan/College Station, Texas 27 Overhead Guide (5) 1 1 1 Overhead Street Name (9) 1 1.4 2 Shoulder Guide (5) 1 1 1 Shoulder Street Name (8) 1 1 1 San Antonio, Texas 36 Overhead Guide (16) 1 1.4 3 Overhead Street Name (13) 1 1.6 3 Shoulder Guide (5) 1 1.4 2 Shoulder Street Name (2) 1 1 1 Orlando, Florida 40 Overhead Guide (11) 1 1.7 3 Overhead Street Name (15) 1 2.5 4 Shoulder Guide (5) 2 2.4 4 Shoulder Street Name (9) 1 1.3 3 Table 11. Signs and complexity ratings by study location. Sign Type, Lighting Location Number of Signs Legend Luminance Legend Height (in.) Avg. Complexity Median Mean St. Dev. Overhead Guide Signs Unlit B/CS 5 11.5 14.0 10.6 16 0.8 Orlando 4 10.8 16.1 18.9 18 1.2 San Antonio 16 8.3 11.4 7.6 16 1.3 Lit Orlando 7 10.9 10.8 1.5 18 2.0 Shoulder Guide Signs B/CS 5 33.3 31.4 12.9 16 0.7 Orlando 5 8.8 13.0 11.0 11–18 2.5 San Antonio 5 37.9 31.1 14.3 8–16 1.2 Overhead Street Name Signs Unlit B/CS 3 10.2 10.9 4.1 9 1.3 Orlando 7 4.5 6.0 4.5 4–9 2.4 San Antonio 10 2.8 3.0 1.3 6–8 1.2 Lit B/CS 6 42.9 41.8 2.3 9 1.4 Orlando 8 42.6 40.8 4.7 9 2.8 San Antonio 3 27.1 28.5 3.7 8 2.0 Shoulder Street Name Signs B/CS 8 2.5 4.8 4.7 6 0.4 Orlando 9 7.8 10.1 7.9 4–9 0.9 San Antonio 2 4.4 4.4 4.3 4–6 0.6 Table 12. Summary of open-road sign data.

31 height when evaluating where drivers were able to correctly read the sign. Average recognition indices of the observations in the open-road study are provided in Table 13. The values are split by sign type and study location. Recognition indices in Orlando tend to be lower than in the other two cities (except for shoulder-mounted street name signs). The recognition indices for guide signs are consistently greater than those for street name signs. Figure 22 shows the distribution of all recognition indices from the study. This distribution is similar in shape to the dis- tribution of legibility distances from the closed-course study in Figure 13. The data are right skewed. While the median and mean values of legibility index from the closed-course study were approximately 44 ft/in., the median and mean values of recognition index from the open-road study were approxi- mately 40 and 42 ft/in., respectively. Under normal conditions, recognition distances are expected to be greater than legibility distances. In fact, previous studies that have measured both have observed recognition distances to be between 1.2 and 1.8 times greater than legibility distances (74–81). The leg- ibility and recognition distances (or indices) from the closed- course and open-road studies are not comparable because the open-road study included signs other than just overhead guide signs, involved a driving task that was more difficult because of the increased work load and increased visual complexity, had more complicated sign legends, and included signs that were not constructed of new materials. The dependent variable analyzed in the open-road study was the recognition index. The values of complexity, lumi- nance, and recognition index in Tables 11 through 13 suggest that some differences in those effects can be attributed to the location (city) and the type of sign viewed. The analysis pre- sented below appropriately includes independent categorical variables for the city and sign type and an interaction term for the combination of city and sign type. The covariates of inter- est were the rated complexity of the visual scene in which the target sign was placed and the luminance of the legend, each based on a photometric image of the sign taken at a dis- tance representing 40 ft from the sign for each inch of legend uppercase letter height. Analysis While the analyses of the closed-course test described in Chapter 3 focused on the individual categorical factors (type of sheeting, type and intensity of sign lighting, and use of street lighting), in addition to the luminance of the sign and Weber contrast as continuous variables, the conditions of the open-road study were not controlled enough to consider individual factors such as street lighting or sheeting type. The recognition index in the open-road study was analyzed with multivariate regression models that used independent vari- ables for luminance, contrast, visual complexity, sign type, and city. The values of luminance and recognition index in Tables 12 and 13 suggest that some of their differences may be attributed to the city and type of sign viewed with possible interactions. The regression models consequently included variables for sign type, city, and their interactions. The dataset contained more than 1,500 total observations of participants reading guide and street name signs. Even though a categorical factor such as use of sign light- ing could have been used, the models included luminance as a covariate instead because the lit overhead guide signs in Sign Type B/CS San Antonio Orlando Overhead Guide Signs 51.7 59.6 41.7 Overhead Street Name Signs 36.2 40.7 25.3 Shoulder Guide Signs 56.4 47.8 39.9 Shoulder Street Name Signs 23.7 26.9 29.8 Table 13. Mean recognition index (ft/in.) by sign type and city. 1.3% 11.9% 17.4% 20.1% 17.8% 14.3% 8.5% 4.8% 2.7% 0.9% 0.2% 0% 5% 10% 15% 20% 25% 10 20 30 40 50 60 70 80 90 100 110 Fr eq ue nc y Recognition Index (ft/in.) Figure 22. Distribution of recognition index from all observations in the open-road study.

32 Orlando had relatively low values of luminance as a result of low retroreflective sheeting (which can be seen in Figure 21). A variable for sign lighting would then poorly indicate how well sign lighting could illuminate signs constructed of newer retroreflective products. The use of luminance as a continu- ous variable is a way to fairly represent the visibility of a sign from the perspective of the driver. Each participant was included as a random effect in the model. This accounted for differences in their visual acuity, driving behavior, or other characteristics unique to each par- ticipant that might affect how well they viewed and read the guide and street name signs. The age group (younger or older) was included to investigate how some differences in recogni- tion index may be attributed to age of the participant, as was done in the closed-course study. Several regression models were created to investigate how luminance, contrast, and complexity of the visual scene influ- ence driver recognition index. A final model was selected from a stepwise process to ensure all fixed effects were significant based on a 95 percent confidence interval (a = 0.05). Weber contrast was not a significant effect. The effects of the final model are identified in Table 14. The resultant model is pre- sented as Equation 2. 38.9 (Eq. 2) 1 2 I AgeGroup City Sign City Sign Complexity Luminance R = + + + + × × β × + β × where IR is the recognition index (ft/in.), AgeGroup represents the age group (younger or older) of the driver, City is the location (Bryan/College Station, San Antonio, or Orlando) where the data were collected, Sign Type is the type of guide or street name sign, Complexity is the rated complexity value of the visual scene where the sign was located, Luminance is the luminance (cd/m2) of the legend, and b1 and b2 are the coefficients of the variables for complexity and legend lumi- nance, respectively. The values for Equation 2 are provided in Table 15. Figure 23 shows a plot of each actual recognition index from the study with the predicted value from the regression model. Despite the broad distribution of residuals (normally distributed, as shown in Figure 24), the consistent matching of the data to the line in the actual-by-predicted plot suggests it is a strong model. Discussion of Model Parameters The following observations can be made from the values for Equation 2 shown in Table 15: • The difference in mean recognition index from the group of younger drivers to older drivers is 7.80 ft/in. Parameter F Ratio Pr > F Age Group 15.13 0.0002 City 4.20 0.0182 Sign Type 212.03 <0.0001 City × Sign Type 20.56 <0.0001 Complexity 13.50 0.0002 Legend Luminance 57.22 <0.0001 Table 14. Main effects of regression model. Parameter Value Intercept 38.9 Age Group Younger 3.90 Older −3.90 City Bryan/College Station 1.10 Orlando −3.94 San Antonio 2.84 Sign Type Overhead Guide Sign 11.06 Overhead Street Name Sign −4.45 Shoulder Guide Sign 5.90 Shoulder Street Name Sign −12.51 City × Sign Type Bryan/College Station Overhead Guide Sign −2.43 Overhead Street Name Sign −0.56 Shoulder Guide Sign 6.05 Shoulder Street Name Sign −3.06 Orlando Overhead Guide Sign −4.03 Overhead Street Name Sign −3.35 Shoulder Guide Sign 0.63 Shoulder Street Name Sign 7.02 San Antonio Overhead Guide Sign 6.74 Overhead Street Name Sign 3.91 Shoulder Guide Sign −6.68 Shoulder Street Name Sign −3.96 β1 (Coefficient for Complexity) −1.61 β2 (Coefficient for Luminance) 0.30 Table 15. Parameters from Equation 2. 0 20 40 60 80 100 120 0 20 40 60 80 Ac tu al R ec og ni tio n In de x (ft/ in. ) Predicted Recognition Index (ft/in.) y=x Figure 23. Actual-by-predicted plot for model of recognition index.

33 • Recognition indices in Orlando were the lowest of the three cities based on the main effect of City without any inter action. San Antonio had the greatest recognition indices. • Based on the main effects disregarding interactions with City, overhead guide signs resulted in the greatest recogni- tion indices of the four sign types, approximately 5 ft/in. greater than the recognition indices of comparable shoulder- mounted guide signs. Overhead street name signs had the third highest recognition index. The smallest mean recogni- tion indices were observed with shoulder street name signs. The large differences in mean recognition index from one type of sign to another suggests that there are specific features of the sign types, such as placement location, that influence how well drivers can read the signs despite already account- ing for the size of the legend. • The different values for the interaction of Sign Type and City suggest that the mean recognition index for each type of sign may vary even by the city where data were collected. • The expected recognition index decreases as the visual scene becomes more complex. The effect is estimated to be approximately 1.6 ft/in. for every increase in complexity from 1 through 5. • The expected recognition index increases as the luminance of the legend increases. The effect is approximately 0.3 ft/in. for every 1 cd/m2 increase in luminance. The categorical effects involving the city and type of sign are included in the model to adjust for differences between the types of signs investigated and the location of the study. Although differences in the two age groups of the participants can be accounted for in random effects for each participant, their inclusion in the model provides insight into how recog- nition index may decrease with age. There was no significant interaction between age group and any other effect. The key finding of the open-road study is the relation- ships between recognition index, visual complexity, and leg- end luminance. The model indicates that recognition index decreases as visual complexity increases. Meanwhile, recog- nition index increases as legend luminance increases. These opposing effects suggest that increases in legend luminance can counteract the effects of increased visual complexity. Based on the ratio of the coefficients for the effects, a 5.6 cd/m2 increase in legend luminance counteracts a unit increase in visual com- plexity. There is no significant interaction between visual com- plexity and legend luminance that affects recognition index. The relationships between recognition index, visual com- plexity, and legend luminance in Equation 2 are linear. Each unit increase in visual complexity or luminance produces a constant and proportional change in recognition index. It is possible that one or both of the true relationships, however, is not linear. Perhaps the influence of luminance on recognition can be more accurately represented by a logarithmic relation- ship, in common with how other sensory inputs are perceived (such as sound intensity measured in decibels). Models with nonlinear relationships were investigated; however, the non- linear relationships provided similar confidence but at the cost of added difficultly for interpretation and application. Conclusions The open-road study focused on how legend luminance and visual complexity of the nighttime scene affect the abil- ity of drivers to successfully recognize specific information from overhead signs. While the method of conducting this experiment on the open road had similarities with the closed- course study described in Chapter 3, there were some key dif- ferences that controlled how the data were analyzed. One was the sizes of the sign legends, which varied because of different sign types and agency policies. Recognition index, which nor- 0 50 100 150 200 250 300 -40 -35 -30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30 35 40 45 Fr eq ue nc y Residual Value (Difference Between Actual and Predicted Values) Figure 24. The model residuals have an approximately normal distribution.

34 malizes the recognition distance by the letter height, was used as the dependent variable. Another difference with the open- road study was the diversity in the sign lighting and sheet- ing materials used. In the field, there were many varieties of sign lighting and sheeting materials experienced, even within a single jurisdiction. The analyses accordingly focused on factors of luminance, contrast, and visual complexity (with categorical variables of sign type and city). The final difference between the two studies was the lack of control for the road- way environment or visual scene in which the study signs were located. One of the principal hypotheses of the open-road study was that the ability of drivers to recognize a destina- tion on an overhead sign is reduced in visually complex and cluttered environments. A complexity level, based on results of computerized image processing and a survey of study par- ticipants, was evaluated as a model covariate to identify its effect on recognition index. The statistical analyses show that recognition index decreases as the visual complexity of the nighttime scene increases. While the visual complexity measure was designed to char- acterize the roadway conditions based on the perspective of the driver, visual complexity may also reflect the operating conditions of the roadway and then, indirectly, the resulting workload of the driver. The implication is that the effect of visual complexity that shows a reduction in recognition index is likely a result of driving under greater workload, where the demands of the driving task interfere with the ability to focus on the study task of reading the target sign. In contrast to the negative effects of visual complex- ity, increased sign legend luminance was found to result in increased recognition index. The study completed on the closed course (described in Chapter 3) in an environment that was not visually complex or demanding in terms of driver workload produced no relationship between luminance and legibility. However, on the open road, where the visual com- plexity of the driving task was higher from both a visual and cognitive perspective, increased legend luminance was found to improve the nighttime driver’s ability to read a sign legend.