Evaluation of The Impact of Spectral Power Distribution on Driver Performance

CHAPTER 8. Final Performance Experiment

Introduction

The purpose of the final experiment for this project was to further investigate variables of interest identified in the previous spectral effects experiments. This experiment melded many factors used in the previous experiments so the combination of their effects could be explored. The configuration of the MPI system performance experiment described in chapter 5 was incorporated to determine the impact on pedestrian visibility at various off-axis positions. The overhead lighting levels incorporated into this study were also used in the mesopic modeling experiment as described in chapter 7. These levels were first researched in the overhead-lighting level experiment described in chapter 6. The impact these factors had in previous experiments were more closely examined in this final performance experiment.

There were no significant main effects for the MPI system configuration on pedestrian-detection distances found in the MPI system performance experiment, but there were significant two-way interactions involving the MPI system. To gather more data on MPI system performance, it was included in this final experiment, but only two configurations were used to reduce the number of variables and test the most likely MPI system configurations: the MPI system off and the MPI system tracking a roadside pedestrian.

The mesopic modeling experiment found mixed results for overhead lighting type. Overhead-lighting type significantly affected threshold contrast in the static portion of the experiment, but as a main effect, it did not significantly affect detection distance. Thus, this final experiment collected more data using the three overhead-lighting types used in previous experiments.

Although reducing the level of overhead lighting results in energy savings, this goal is not acceptable if it compromises visibility. A combination of overhead-lighting level and pavement type affects adaptation luminance, which in turn affects the eye’s behavior in the mesopic range. In this experiment, as in the mesopic modeling experiment, the experiment was conducted with five distinct adaptation luminances to determine the points at which visibility might be compromised.

Driver behavior differs at varying driving speeds, possibly resulting in different levels of visual performance. Thus, this experiment included two different speeds common to roadways, 56 km/h and 80 km/h (35 and 50 mi/h).

Pedestrians were employed as the detection targets for this experiment because the current state of machine-vision technology limits an MPI system’s ability to detect and highlight small targets. The pedestrians were placed at various distances from the roadway. They wore different colors and appeared on both the left and right sides of the road. These conditions were selected to mimic naturalistic conditions, to reduce the chance that a participant might predict pedestrian position, and to introduce the possibility of detection at multiple eccentricities because eccentricity affects mesopic visual performance. Eccentricity was not controlled in this experiment, however, because participants were not instructed to keep their eyes focused on the roadway. To create the possibility for peripheral detection, pedestrians were placed at various offsets with respect to the driving lane. Participants could scan for pedestrians, and detection could take place either in the fovea or in the periphery.

Research Objectives

The research objectives of the final performance experiment were the same as those of the overall project. They included evaluating the following:

• Impact of the spectra of overhead-lighting systems on driver visual performance.
• Interaction of vehicle headlamps and overhead lighting in terms of object visibility.
• Applicability of mesopic models and scaling factors in a roadway lighting design.
• Impact of a peripheral illumination system on driver visual performance.

Experimental Design

A 2 by 3 by 5 by 2 by 2 by 4 by 2 by 3 mixed-factors experiment was designed to measure the effect of age, overhead lighting type, adaptation luminance, MPI system configuration, vehicle speed, visual angle, pedestrian position, and pedestrian clothing color on pedestrian-detection and color-recognition distances. The variables used in the experiment are listed in table 34, through table 36.

Table 34. Final experiment—independent variables and values.

Table 35. Final experiment—covariate and measurement method.

Table 36. Final experiment—dependent variables and measurement method.

Independent Variables

Age

Participants were divided into the same age groups as previous experiments: drivers (25–35years old) and older drivers (65 years old and older).

Overhead Lighting Type

All three overhead-lighting systems were used.

Adaptation Luminance

Adaptation luminance was the luminance, viewed from inside the test vehicle, for a given combination of overhead-lighting level and pavement type. The same design was used here as in the mesopic modeling experiment, and adaptation luminances are reported in table 37.

Table 37. Final performance experiment—adaptation luminance per overhead lighting level and pavement type.

MPI System Configuration

The mockup MPI system was configured to behave in two ways: off, with headlamps aimed ahead, like in a normal vehicle; and tracking, with headlamps illuminating a pedestrian from about 183 m (600 ft) away, and then swiveling the headlamp to keep the pedestrian in the beam as the vehicle approached. The configurations are illustrated in figure 59.

Speed

The participant drove the vehicle at 56 and 80 km/h (35 and 50 mi/h).

Pedestrian Offset

The eccentricity at which participants detect pedestrians for dynamic experiments cannot be fixed, because as the driver approaches the pedestrian, the eccentricity is constantly changing. Also, for this experiment, participants were not instructed to focus along the roadway. To create the possibility for peripheral detection, pedestrians were placed at four offsets with respect to the travel lane. To determine the pedestrian positions, a theoretical detection distance was fixed at 83m (277 ft), because based on a 1-degree downward viewing angle with a vehicle height of 1.45 m (4.76 ft), this is the resulting distance.⁽⁵⁾ Four offsets were calculated based on detection from that distance—3.0, 7.7, 8.9, and 21.0 m (9.8, 25, 29, and 69 ft)—and are described in figure 128.

1 m = 3.3 ft

Figure 128. Diagram. Final performance experiment—pedestrian positions and offsets from the roadway.

All pedestrian positions were in the lighted section of the Smart Road, and all pedestrian positions had VI on the pedestrian’s face as similar as possible to each other within a specified lighting condition. VI levels are listed in table 38.

Table 38. Final performance experiment—VI for all pedestrian positions, overhead lighting levels, and visual angles.

Pedestrian on Left or Right

The pedestrian stood on the left or right of the road with respect to the participant vehicle. Pedestrians at 3.0 m (9.8 ft) offset were located on both sides of the road. Pedestrians at 7.7 m (25 ft) offset were located only on the left side of the road, and pedestrians at 8.9 and 21.0 m (29and 69 ft) offset were located only on the right side of the road, as shown in figure 128.

Pedestrian Clothing Color

Pedestrians wore clothing in three colors: gray, red, and blue, as described in chapter 3.

Covariate

Contrast

The luminance of the pedestrians and the background behind them was measured using the ProMetric® luminance system from 83 m (277 ft) up the road from the pedestrian, the same distance at which RP-8 advises roadway lighting designers to measure roadway luminance from a luminaire.⁽⁵⁾ The Weber contrast was calculated from that measurement.

Dependent Variables

Detection and color-recognition distances were the dependent variables. Orientation-recognition distance was not measured, and pedestrians all faced toward the roadway.

Methods

Facilities and Equipment

This experiment was conducted on the Virginia Smart Road using the same test vehicles as in previous experiments. The headlamps were the HID headlamps installed on the test vehicles for the previous experiments in this project but with low-intensity neutral-density filter with a CCT of 3,530K and transmittance of .3413.

Participants

Participants were recruited and screened as described in Chapter 3; however, for this experiment, they were not tested for UFOV.

A total of 36 participants performed the experiment, 18 older and 18 younger. Within each age group, participants were divided equally by gender. Mean and standard deviation of participant age, visual acuity, mesopic visual acuity, and low contrast visual acuity are listed in table 39.

Table 39. Final performance experiment participant characteristics.

Procedure

Participants were recruited, screened, and directed to the Smart Road as described in chapter 3. Each participant attended three experimental sessions, one for each overhead-lighting type. Each session consisted of 1 practice run and 12 experimental runs up and down the Smart Road, for a total of 24 trials for each lighting type. Lighting level was changed twice during each session, so all three intensities were tested. Two participants in two vehicles typically completed the experiment at one time.

During the experiment, on-road experimenters adjusted the overhead lighting level and pedestrian positions for the each experimental run. In-vehicle experimenters directed participants to drive at the speed called for by the protocol and recorded detection distances.

Data Analysis

Detection and Color Recognition Distances

After video data reduction, an ANOVA was used to determine whether the independent variables significantly affected pedestrian-detection and color-recognition distances. When results were significant, Tukey HSD tests were performed to determine which factors differed from each other. When Tukey HSD tests are reported on charts, data points sharing a letter do not significantly differ from each other.

For missed detections, Fisher’s Exact Test was conducted to determine the degree of difference between two binomial variables. The results of the test are reported as the probability that onevariable occurs over another variable.

This experiment focused mainly on gray-target detection; colored targets were not distributed evenly among the visual angles. Therefore, although data analysis was performed on all target colors, it was repeated only for the gray targets.

Contrast

Weber contrast was calculated for the pedestrians the same way it was calculated for the other experiments in this project. An ANCOVA using contrast as a covariate was performed to determine the relationship between contrast and detection distance. Contrast was not analyzed with respect to color-recognition distance. Detection is more crucial for driving safety than color recognition. Only luminance contrast, not color contrast, was measured.

Contrast and detection-distance data were analyzed separately for when the MPI system was off and when it was on, but results indicated the MPI system introduced a great deal of variation in the data—system-on data and system-off data were not necessarily comparable. In addition, it was not possible to calculate the contrast for the instant a participant detected a pedestrian, because it was calculated from luminance measurements taken 83 m (277 ft) along the road from the pedestrian. Actual detection occurred at between 124 and 97.4 m (407 and 320 ft), introducing error in the contrast measurement.

Results

Main Effects

Main effects on detection distance and color-recognition distance occurred for many of the same independent variables. Significant main effects are listed in table 40 with results explained in detail following the table.

Table 40. Final performance experiment results.

Table 41. Final performance experiment significant main effects, gray only.

Adaptation Luminance

The results for adaptation luminance’s effect on detection and color-recognition distances are shown in figure 129. When all pedestrian clothing colors were analyzed, it was found adaptation luminance significantly affected detection distance, with mean detection distance at 0.1 cd/m² (0.03 fL) shorter (M = 95.7 m (314 ft)) than at the other adaptation luminances (all more than 109 m (358 ft)). The same effect was seen in the previous experiment. Adaptation luminance also significantly affected color-recognition distance, following the same trend as described for detection distance but to a lesser extent. This could be because of poor target contrast at that luminance, a condition described in more detail in the contrast section of the results.

When only gray-clothed pedestrians were analyzed, it was found that adaptation luminance also significantly affected detection distance, with a general trend of higher adaptation luminances having longer detection distances.

1 cd/m² = 0.3 fL
1 m = 3.3 ft

Figure 129. Graph. Final performance experiment—detection distance and color-recognition distance by adaptation luminance.

Age

Age significantly affected pedestrian-detection distance when data for only gray-clothed pedestrians were analyzed. The mean detection distance for older participants (M = 107 m (351 ft)) was significantly shorter than that for younger participants (M = 135 m (440 ft)).

The Snellen visual acuity test performed during participant screenings found that older participants had poorer visual acuity than younger participants. While normal vision is 20/20, younger participants had average visual acuity of 20/17.2, while older participants had an average visual acuity of 20/21.6. Older participants also had poorer contrast sensitivity than younger participants. When the contrast sensitivity exam was administered with lighter-colored gray letters on the Snellen eye chart, younger participants had an average acuity of 20/20.8, while the older participants had an average of 20/27.8. These two exams were performed in photopic conditions. A third Snellen acuity test was performed in mesopic conditions; for that exam, younger participants had an average visual acuity of 20/25.7, while older participants had an average visual acuity of 20/38.1. As the difficulty of the Snellen eye exams increased, from normal-to-low contrast to mesopic conditions, older participants’ visual acuity decreased when compared with the younger participants. That difference in visual acuity was reflected in the gray-only target detection distances because the colored targets were distributed unevenly among the visual angles, introducing variations in the detection distances not caused by color alone.

Pedestrian Clothing Color

Mean detection distances were significantly different among the three pedestrian clothing colors, with gray (M = 123 m (404 ft)) seen from farther away than red (M = 87.3 m (286 ft)) and blue (M = 79.4 m (260 ft)). Mean color-recognition distances did not differ significantly by pedestrian clothing color. Pedestrian clothing color results are shown in figure 130.

1 m = 3.3 ft

Figure 130. Chart. Final performance experiment—detection and color-recognition distances by pedestrian clothing color.

The experiment was designed so that pedestrians wearing gray were seen 10 times more frequently than those wearing red or blue, and the visual angle was not divided evenly among the colors. Therefore, definitive results regarding pedestrian clothing color cannot be obtained.

Offset

For all pedestrian clothing colors, detection distance differed significantly with offset. As offset increased, detection distance decreased. A Tukey HSD test found that the only detection distances that differed significantly from each other were those at 3.0 m (9.8 ft) offset (M = 124 m (407 ft)) and 21.0 m (69 ft) offset (M = 97.4 m (320 ft)). This result was expected, because visual acuity decreases as offset increases and objects are viewed more in the eye’s periphery. Figure 131 illustrates the results.

Color-recognition distance differed significantly with offset but followed a different trend (figure 131). Color-recognition distance was shortest at 21.0 m (69 ft) offset (M = 58.8 m (193ft)) and longest at 8.9 m (29 ft) offset (M = 76.6 m (251 ft)), but the standard deviation of the color-recognition data, between 47.5 and 68.9 m (156 and 225 ft) depending on offset, warrants caution when attempting to draw meaning from the results.

1 m = 3.3 ft

Figure 131. Chart. Final performance experiment—detection distance and color-recognition distance versus offset for all pedestrian clothing colors.

MPI System Configuration

The MPI system configuration did not significantly affect detection distance but did significantly affect the percent of detections, as tested with a Fisher’s Exact Test. The probability that MPI-off resulted in significantly more misses than MPI-on was significant. With the MPI system on, drivers either missed or did not detect 24 percent of the pedestrians. With the MPI system off, drivers missed 38 percent of pedestrians.

Two-Way Interactions

There were a number of significant and marginally significant two-way effects on detection distance and percent of missed detections.

Offset and Adaptation Luminance

For all pedestrian clothing colors, the combination of adaptation luminance and offset affected detection distance (figure 132). At an adaptation luminance of 0.07 cd/m² (0.020 fL), detection distances decreased with increasing offset. This is expected, given that visual acuity decreases as eccentricity increases, and if drivers’ eyes were focused on the roadway, higher offset would correspond to higher eccentricity. At higher adaptation luminance levels, however, the relationship between offset and detection distance changed; detection distance decreased with increasing offset, but the trend was less clear and more general. A similar trend was seen for only gray-clothed pedestrians, as shown in figure 133.

1 cd/m² = 0.3 fL
1 m = 3.3 ft

Figure 132. Chart. Final performance experiment—detection distance by adaptation luminance and offset for all clothing colors.

1 cd/m² = 0.3 fL
1 m = 3.3 ft

Figure 133. Chart. Final performance experiment—detection distance by adaptation luminance and offset for gray-clothed pedestrians only.

Offset and Age

The number of missed detections at each location indicated that there was a much greater probability for missed detections at 21 m (69 ft) offset than at the other offset distances (figure 134). The effect was greater for older participants, who missed 61 percent of pedestrians at 21.0m (69 ft) offset, than it was for younger participants, who missed 29 percent of pedestrians at 21.0 m (69 ft) offset (Fisher’s Exact Test, p ≤ 0.0001).

1 m = 3.3 ft

Figure 134. Chart. Final performance experiment—missed detections by age and offset.

Offset and Overhead Lighting Type

For gray-clothed pedestrians, the interaction effect between offset and overhead-lighting type was not significant; however, there were differences in detection distances among the overhead-lighting types that depended on offset. At 3.0 and 7.7 m (9.8 and 25 ft) offset, the overhead-lighting types had similar detection distances. At 8.9 and 21.0 m (29 and 69 ft) offset, the 6,000‑K LED lighting had greater detection distances, possibly because the 6,000-K LED lighting’s spectral distribution is more efficient in the mesopic range, and mesopic effects are only seen in the periphery. At 21.0 m (69 ft) offset, the lighting types’ detection distances increased with increasing color temperature. The results are illustrated in figure 135.

1 m = 3.3 ft

Figure 135. Chart. Final performance experiment—detection distance by offset and overhead-lighting type for gray-clothed pedestrians.

Offset and MPI System Configuration

Offset and MPI system did not have an interaction effect on detection distance but did have an interaction effect on missed detections. The MPI system was not used for pedestrians at the 3.0m (9.8 ft) offset. When detection rates were analyzed without the 3.0 m (9.8 ft) offset data, it was found that when the MPI system was on, participants missed the pedestrians at 7.7 and 8.9 m (25 and 29 ft) offsets more than when it was off (figure 136). Conversely, for the pedestrians at 21.0 m (69 ft) offset, participants missed more pedestrians with the MPI system off than with it on (Fisher’s Exact Test, p ≤ 0.0001).

Similarly to its main effect, the MPI system’s on or off status did not affect detection distance but did affect detection rate; however, it did so differently for different offsets. It was most beneficial in increasing detection rates for the pedestrian at 21.0 m (69 ft) offset. It could have had a slight negative effect on detection rates for the pedestrians at 7.7 and 8.8 m (25 and 29 ft) offsets because the MPI system swiveling beam could have caused an extreme change in contrast. At the 7.7 m (25 ft) offset, missed detections were significantly greater with the MPI system on than off (Fisher’s Exact Test, p = 0.0226). At the 8.9 m (29 ft) offset, the difference was not significant.

1 m = 3.3 ft

Figure 136. Chart. Final performance experiment—missed detections by MPI system configuration and offset.

Offset and Speed

For all clothing colors, participants driving 80 km/h (50 mi/h) detected the pedestrians from slightly farther away (M = 111 m (364 ft)) than those driving 56 km/h (35 mi/h) (M = 119 m (390 ft)), but the main effect was not significant. However, a two-way interaction between speed and offset found that detection distances were longer for higher speeds at 3.0, 8.9, and 21.0 m (9.8, 29, and 69 ft) offsets but not at the 7.7 m (25 ft) offset (figure 137).

1 m = 3.3 ft
1 km/h = 0.62 mi/h

Figure 137. Graph. Final performance experiment—detection distance by speed and offset for all clothing colors.

Some explanations for these results could stem from roadway design, driving speed, and scanning behavior. At faster speeds, drivers are likely more vigilant and might have different scanning behavior. Drivers at higher speeds might be more vigilant, scanning might be restricted to within a few degrees of the roadway on the left but cover a broader angle on the right, or rely more on periphery. Therefore, at higher speeds, more-vigilant drivers would detect pedestrians from farther away within their scanning region. The pedestrians at 7.7 m (25 ft) offset, corresponding to 5 degrees, were only on the left side of the road, and the location of the vehicle A pillar might have hindered scanning patterns and limited detection distances at that visual angle for both speeds.

A similar effect was seen when only data for gray-clad pedestrians were analyzed. Detection distances were longer for higher speeds at 3.0 and 21.0 m (9.8 and 69 ft) offsets (2 and 14degrees at 83 m (277 ft)), but detection distances were shorter for higher speeds at 7.7 and 8.9m (25 and 29 ft) offsets (5 and 6 degrees at 83 m (277 ft)) (figure 138).

1 m = 3.3 ft
1 km/h = 0.62 mi/h

Figure 138. Graph. Final performance experiment—detection distance by speed and offset for gray only.

Overhead Lighting Type and Adaptation Luminance

To isolate the effects of overhead lighting and adaptation luminance, figure 139 shows their combined effect on detection distance in cases when the MPI system was off. Figure 135 shows that overhead lighting with a higher CCT and with an SPD more efficient in the mesopic range had longer detection distances but only for larger eccentricities, where mesopic effects are expected. One would also expect higher adaptation luminances to correlate with longer detection distances for large-object detection because the general environment is brighter. However, that effect was only seen for the LED lighting; at adaptation luminances of 0.07 and 0.1 cd/m² (0.020 and 0.03 fL), detection distances for the LED lighting types were shorter than for adaption luminances of 0.2cd/m² (0.06 fL) and higher. There may exist a threshold adaptation luminance, above which brighter lighting does not significantly increase the visibility of large objects such as pedestrians, thus creating a limited return to additional lighting. The interaction of adaptation luminance and color might also explain the difference because overhead-lighting type and adaptation luminance did not significantly affect detection distances for gray-clothed pedestrians.

1 cd/m² = 0.3 fL
1 m = 3.3 ft

Figure 139. Chart. Final performance experiment—detection distance by adaptation luminance and overhead-lighting type with the MPI system off.

Contrast and Detection Distance

Weber contrast was calculated from luminance measurements taken 83 m (277 ft) along the road from the pedestrian positions. An ANCOVA was performed using contrast as a covariate. Contrast and detection-distance data were only analyzed for gray pedestrians because only luminance contrast, not color contrast, was measured. The ANCOVA was only performed for detection distance. The results are listed in table 42.

Table 42. Final performance experiment results for Weber contrast and detection distance for gray pedestrians only.

Adaptation Luminance

The detection distance at 0.1 cd/m² (0.03 fL) was lower than at the other adaptation luminance levels, and the Weber contrast results could explain why. At 0.1 cd/m² (0.03 fL), the pedestrian was seen in positive contrast, but at all the other adaptation luminance levels, the contrast was negative, as illustrated in figure 140. At some point, when the vehicle approached the pedestrian, the pedestrian contrast passed from negative to positive. Detection is much more difficult as contrast passes through zero, causing longer detection distances.

1 cd/m² = 0.3 fL
1 m = 3.3 ft

Figure 140. Graph. Final performance experiment—detection distance and Weber contrast by adaptation luminance.

MPI and Adaptation Luminance

When the MPI system was on, detection distances were shorter, albeit not statistically significantly. That effect was greatest at 0.1 cd/m² (0.03 fL). At most adaptation luminance levels, the MPI-on condition caused the Weber contrast to be closer to zero because the headlamps illuminated the pedestrian, bringing him or her closer to positive contrast, as shown in figure 141. Thus, while the MPI system illuminated the pedestrian, it also decreased contrast, making detection more difficult. The MPI system could have also distracted the participant, negatively affecting detection distances.

1 cd/m² = 0.3 fL
1 m = 3.3 ft

Figure 141. Graph. Detection distance and Weber contrast by adaptation luminance for MPI system on and off.

Offset

As offset increased, contrast approached zero and detection distance decreased (figure 142). There was a sharp increase in contrast, and contrast passed through 0, 8.9, and 21.0 m (29 and 69ft) offsets. There was also a sharp decrease in detection distance between 8.9 and 21.0 (29 and 69 ft) offsets. Contrast polarity factor (C_FP), described by Adrian, states that objects in negative contrast are easier to detect than those in positive contrast.⁽⁵⁴⁾ In this experiment, the contrast at the 21.0 m (69 ft) offset was positive, but at the other offsets, the contrast was negative. That could help explain why detection distances were so much shorter at the 21.0 m (69 ft) offset than at the other offsets, even though the contrast at 21.0 m (69 ft) offset was close to 0.1.

Probability summation, or the probability of detecting an object with both eyes, cannot be factored into the visibility of the targets at the various offsets because of this study’s lack of control over the participants’ eye gaze behavior, and offsets did not necessarily correspond to eccentricity.

1 m = 3.3 ft

Figure 142. Graph. Final performance experiment—detection distance and Weber contrast by offset.

DISCUSSION

The main purpose of this experiment was to further evaluate the variables of interest identified in previous experiments in this project.

One of the outcomes of the scoping experiment was that more information was needed on the effect of overhead-lighting type and level on peripheral visibility, and a project objective was to address the effect of the SPD of overhead lighting on driver visual performance. The mesopic modeling experiment found that the spectral distribution of overhead lighting affected detection distances but only at higher eccentricities. This experiment attempted to extend the investigation to determine the effects of overhead lighting’s SPD on visibility while driving.

The final performance experiment found no significant differences in detection and color-recognition distances between overhead-lighting types at any offset, although results showed a trend toward longer detection distances with 6,000-K LED lighting at greater offsets. The results do not strongly support a spectral effect of overhead lighting on mesopic visibility in the periphery. That result, different from the one in the mesopic modeling experiment where eccentricity was fixed, means that drivers depended on glance patterns to scan their driving environment and detect peripheral objects. Therefore, they likely detected objects in the fovea, where mesopic effects do not occur.

MPI

Evaluating the performance of drivers when using a peripheral illumination system was a project objective. This experiment found that the MPI system configuration did not significantly affect detection distance. This is consistent with findings from the spectral interaction experiment, where the headlamp’s effect on visibility was much less than that of the overhead-lighting level. In the final performance experiment, most detection occurred at distances greater than the 75 to 90 m (246 to 295 ft) window where the headlamps first illuminated the pedestrian. Therefore, in most cases, the MPI system highlighted the off-axis pedestrians after participants had detected them.

The MPI system configuration did significantly affect the probability that a pedestrian was detected, but the results varied among offsets. For the pedestrian at 21.0 m (69 ft) offset, the MPI system significantly increased detection rates. For the pedestrians at 7.7 and 8.9 m (25 and 29 ft) offsets, the MPI system slightly decreased detection rates. Congruent with the results of the spectral interaction experiment, at the 21.0 m (25 ft) offset location, luminance from overhead lighting was low, and headlamps were the primary light sources driving detection. At the 7.7 and 8.9 m (25 and 29 ft) offset locations, the pedestrians were largely illuminated by overhead lighting and were in negative contrast. When the headlamp light struck them, it reduced their contrast until they shifted into positive contrast. Near-zero contrast conditions reduced visibility and detection distances.

Because the MPI system only increased the detection rate for pedestrians farthest from the roadway, it is important to consider how far an object is from the roadway to evaluate its potential as a hazard. The pedestrian at the highest offset was 21.0 m (69 ft) to the right of the test vehicle (figure 128). If the pedestrian is traveling directly toward the road, and the vehicle is traveling at 80.4 km/h (50 mi/h) and 83 m (272 ft) from the point where the pedestrian would intersect the road, the pedestrian would have to be moving 20.3 km/h (12.6 mi/h) for it to intersect the vehicle and cause a collision. Large animals native to North America, such as deer and black bear, can easily achieve that speed, as can some runners and cyclists.^(98,99)

The MPI system might be distracting to drivers. Results of the MPI system performance experiment indicated the MPI system illuminating an area of road with no pedestrian interfered with drivers detecting pedestrians on the opposite side of the road. This research has shown that when overhead lighting is in place, the MPI system does not significantly alter detection distances, but it did increase the probability of detection, especially at high eccentricity angles. Further research should explore these limits. In addition, driver behavior associated with false positives and other potential errors the system should be further investigated.

Age

Age did not significantly affect detection or color-recognition distances for this group of participants, who were all experienced drivers and comfortable with nighttime driving. The difference in detection distances between the younger and older groups, however, was effectively different at 22 m (72 ft), the difference in safe stopping distances for vehicles traveling 50 and 60km/h (31 and 37 mi/h). Other studies reported in chapter 2 found diminishing visual acuity with age, but peripheral contrast sensitivity diminishes with age more slowly than foveal contrast sensitivity.⁽⁹⁵⁾ If participants initially detected the pedestrians using peripheral vision, that could explain the lack of age effects, similar to the lack of age effects in the mesopic modeling experiment, which ensured peripheral detection.

Color

Gray-clad pedestrians were often detected from farther away than red- or blue-clad pedestrians. Color-recognition distances for the red- and gray-clad pedestrians were longer than for the blue-clad pedestrians. The experimental design, and the fact that color contrast was not measured, meant that it was difficult to determine why those results occurred.

Adaptation Luminance

Although this experiment did not specifically evaluate mesopic models in a driving environment—a project objective—it did address visual performance in the mesopic range. The detection differences among the adaptation luminance levels were in line with Adrian’s model.⁽⁵⁴⁾ As adaptation luminance changed, so did contrast threshold. Comparing VL to Weber contrast (figure 143) showed that as adaptation luminance increased, contrast decreased and VL increased. This is because higher adaptation luminances require higher threshold contrasts for object detection.

1 cd/m² = 0.3 fL

Figure 143. Graph. Final performance experiment—VL and Weber contrast by adaptation luminance.

At the lowest adaptation luminance level, 0.07 cd/m² (0.020 fL) and at 14 degrees (corresponding to a 21.0m (69 ft) offset at 83 m (277 ft) distant), light levels approached the lower end of mesopic vision. The eye is more adapted to darkness and more contrast-sensitive in those conditions, possibly causing the longer detection distances than those at 0.1 cd/m² (0.03fL), where vision is more mesopic.

In general, visual performance depended more on adaptation luminance than on overhead-lighting type. At higher adaptation luminances and closer to the photopic region, overhead-lighting type had even less effect on visual performance. At the lowest adaptation luminance, detection was only about 10 m (30.5 ft) shorter than at higher adaptation levels, likely because a dark-adapted eye is more contrast sensitive. The fact that detection distances were longer at the larger visual angles, where both pedestrians and background were darker, supports that conclusion. The 0.1 cd/m² (0.03 fL) adaptation luminance showed the shortest detection distances and was the only adaptation luminance level where the target had positive contrast. The contrast of that adaptation level probably passed through zero as the vehicle approached the pedestrian, reducing the pedestrian’s visibility.

Offset

There was no difference in detection or color-recognition distances among the different lighting types for the pedestrians stationed closest to the roadway. For pedestrians at 21.0 m (69 ft) offset, the two LED lighting types had longer detection distances than HPS lighting, because the LED lighting had a greater S/P ratio and better SPD in the mesopic range. That result might have occurred only at 21.0 m (69 ft) offset (corresponding to 14-degree eccentricity at 83 m (277 ft) away) because the maximum rod density is at 15 degrees and is where a mesopic effect would most likely occur.⁽¹⁰⁰⁾ However, without restricting the participants’ eye-glance behavior, it is difficult to determine whether detection occurred in the fovea or periphery.

Speed

Speed interacted with visual angle to significantly affect detection distance, with detections longer for 3.0, 8.9, and 21.0 m (9.8, 29, and 69 ft) offsets but not the 7.7 m (25 ft) offset, when the participant was driving faster. This could be because driver scanning behavior is broader when driving slower and more vigilantly and narrower when focused down the road and driving faster. At faster speeds, drivers might also focus more on the roadway and rely more on peripheral cues for object detection on the shoulder.

Conclusions

The four objectives of this project—and of the final performance experiment—were to evaluate the following: (1) impact of the spectra of overhead lighting systems on driver visual performance; (2) interaction of vehicle headlamps and overhead lighting in terms of object visibility; (3)applicability of mesopic models and scaling factors in a roadway lighting design; and (4)impact of a peripheral illumination system on driver visual performance.

First, the impact of spectra of overhead-lighting systems on off-axis visibility was minimal because changes in lighting type did not affect visibility or color recognition significantly. The results of the mesopic modeling experiment indicated significant differences in the overhead-lighting systems were found in the static but not in the dynamic portions of the experiment. The results of this experiment indicate the same is true for pedestrian visibility as it was for targets.

Second, peripheral models were found to be not effective in predicting visual behavior in a driving environment. The inability to control where drivers are looking and what they are attending to at any given moment during a driving task, makes predicting visibility of any roadway object difficult, especially those in low contrast.

Third, this experiment found that the MPI system provides a benefit for off-axis visibility in terms of whether pedestrians are detected or not detected, but the system does not provide a clear benefit of detection distance and may be considered a distraction from the forward roadway.

Other findings were that object size, ambient lighting level, and contrast affected the visibility of gray-clad pedestrians and gray targets in this experiment. The larger an object, the more visible it is, as shown by the VL calculations; pedestrians had higher VLs than targets. Also, the ambient lighting level affects the eye’s adaptation, which in turn affects contrast sensitivity. Findings related to contrast polarity and ambient luminance were that high ambient luminance can make a positively contrasting object difficult to see, and low ambient luminance can make a negatively contrasting object difficult to see. An object with very low contrast will be difficult to see regardless of object size and adaptation level; however, that poor visibility can be mitigated by adjusting object size and adaptation luminance. In addition, a target in negative contrast becomes briefly invisible as it transitions to positive contrast when it starts to be illuminated by headlamps. The effects of higher adaptation levels to brighter environments are particularly noticeable at night with low-contrast objects. Findings related to detail recognition were that orientation recognition typically occurred only within 30 m (100 ft) of the object, requiring the driver to drive slower than 24 km/h (15 mi/h) to be able to stop in time to avoid colliding with the object. Distant objects were more visible without headlamps than with headlamps, likely because the headlamp light caused the eye to adapt to the brighter environment.