Evaluation of The Impact of Spectral Power Distribution on Driver Performance

CHAPTER 5. MPI System Performance Experiment

Introduction

As discussed earlier, one promising technology for increasing pedestrian and object visibility on nighttime roadways is an MPI system that would detect and highlight crash hazards. With this technology, objects located in the areas of a driver’s peripheral vision could be illuminated more effectively than with a traditional headlamp design. With illuminating objects in a driver’s peripheral vision comes the consideration of how a driver’s eyes process headlamp color. It is also worth exploring driver reactions to that technology. For example, it is unknown whether moving headlamps would distract drivers to the extent that visibility would be decreased instead of increased. Therefore, the goal of this experiment was to test the visual performance and extent of distraction of participant drivers using an MPI system to relate such a system to detection and recognition of color located in the periphery of driver’s visual field.

Designing an actual MPI system was outside the scope of this project. However, a mockup MPI system that could not detect pedestrians but instead moved the headlamps to illuminate pedestrians at predetermined locations, was designed, built, and used. The MPI system behaved in two ways, either turning on the headlamp at a set distance and spot-lighting the pedestrian for a short period of time, or swiveling the headlamp to keep the pedestrian in its beam while the vehicle approached. Two headlamp colors were used—the white/yellow and white/blue used in the scoping experiment—but only the high-intensity filters were used. Because small target detection is difficult using machine vision software, MPI systems would likely be designed to illuminate only pedestrians. Thus, this experiment used pedestrians only to test visibility. Pedestrians were placed on both sides of the road and wore two clothing colors.

Detection distance, detection rate, and color-recognition distance quantified visual performance. A major concern with using MPI systems is the extent to which they would distract the driver. Detection distances cannot entirely capture driver distraction, so an eye tracker was used in this experiment to measure the driver’s fixation duration.

Research Objectives

One of this project’s objectives was to evaluate the impact of a peripheral illumination system on driver visual performance. To achieve that objective, the purpose of the MPI system performance experiment was to evaluate the following:

• How different highlighting behaviors of the MPI system affect driver visual performance.
• How the MPI system and overhead lighting interact to affect driver visual performance.
• How different headlamp colors on an MPI system affect driver visual performance.
• How an MPI system affects driver eye-glance behavior.

Experimental Design

A mixed-factors experiment was conducted investigating the effects on participant detection distances and detection rates and on participant eye-glance behavior. Factors considered were participant age, headlamp color, overhead-lighting level, MPI system configuration, pedestrian position, pedestrian clothing color, and whether the pedestrian was illuminated. Variables are listed in table 15 and table 16.

Table 15. MPI system performance experiment independent variables and values.

Table 16. MPI system performance experiment dependent variables and measurement method.

Independent Variables

Age

For the same reason as in the scoping experiment, this experiment included two age groups: younger (25–35) and older (65+).

Headlamp Color

Two headlamp configurations were used in this experiment: white/yellow high intensity and white/blue high intensity. The high-intensity filters were used because they more closely matched the intensity of headlamps on the market.

MPI System Configuration

The purpose of this experiment was to measure driver reactions to a system with moving headlamps that illuminate pedestrians, rather than to develop a full-featured working MPI system. Therefore, the team designed and built a mockup MPI system with headlamps swiveled and aimed according to a predetermined program, rather than by detecting the pedestrians in real time. The headlamps are shown in figure 58.

Figure 58. Photo. MPI system with straight (left) and swiveling (right) headlamp.

Confederate pedestrians stood on the roadway in the positions the headlamps were programmed to illuminate. The mockup MPI system had three configurations. The first was highlighting, in which the system would aim a headlamp at a pedestrian and turn it on momentarily, illuminating the pedestrian for a short period of time. Highlighting occurred when the vehicle was about 46.7m (150 ft) from the pedestrian, and only the headlamp closest to the pedestrian would highlight him or her. The second MPI system configuration was tracking, in which the MPI system turned on the headlamp closest to the pedestrian when the vehicle was about 183 m (600ft) from him or her and then swiveled the headlamp to keep the pedestrian in the beam as the vehicle approached. The last configuration was when the MPI system was off, with the headlamps aimed down the roadway as in a normal vehicle. The three MPI system configurations are illustrated in figure 59. The MPI system illuminated pedestrians on both sides of the road.

Figure 59. Diagram. MPI system performance experiment—MPI system configurations.

Pedestrian Clothing Color

Pedestrians wore blue or red scrubs. The blue color was chosen because both the 6,000-K LED overhead lighting and white/blue headlamps have strong blue components. The red was chosen because the white/yellow headlamps have a strong amber component. Both red and blue are common clothing colors.

Pedestrian Stations

Pedestrians were placed at various locations along the road to create the independent variables described in the following sections. All pedestrian stations were placed away from curves in the road, so line-of-sight would not be shorter than the detection distance. Stations were placed far enough apart so that participants would not detect two pedestrians at two different distances at the same time. VI on the pedestrians’ faces was kept constant at 0.40 lx (.037 fc) for all pedestrian stations. Although the pedestrians’ faces were illuminated at 0.40 lx (0.037 fc), the background behind the pedestrians was necessarily different, creating different contrasts.

Pedestrian on Left or Right

Pedestrians were placed on the left-hand or right-hand side of the roadway with respect to the oncoming vehicle, approximately 12 m (40 ft) away from the shoulder.

Pedestrian in Overhead Lighting

Pedestrians stood at five different stations. Two of those stations were illuminated by overhead lighting, and three were not. The areas with overhead lighting used 6,000-K LED luminaires dimmed to 20 percent. At that dim level, the horizontal illuminance was 1.97 lx (0.183 fc) when measured on the ground, halfway between two luminaires, on the shoulder of the road opposite the luminaires. That dim level also resulted in a VI of 0.40 lx (0.037 fc) on the pedestrian’s face and enabled researchers to consistently illuminate the pedestrians in the sections of the road with overhead lighting.

Pedestrian Illuminated by MPI System

To test whether the MPI system would distract drivers from seeing pedestrians or objects outside of its beam path, some runs were performed with a pedestrian on the side of the road opposite from where the MPI system illuminated, illustrated in figure 60.

Figure 60. Diagram. MPI system performance experiment—pedestrian opposite from area illuminated by MPI system.

Dependent Variables

Detection and color-recognition distances were measured in this experiment, as described in chapter 3. Orientation-recognition distance was not measured, and pedestrians faced the same direction for all runs. Fixation duration was also measured and is described the following subsections.

Fixation Duration

Fixation duration was measured using an Arrington Research ViewPoint EyeTracker® and its accompanying software. The eye tracker was mounted on a pair of clear goggles with twocameras, one under each eye, to determine the direction the pupil was aimed. A third camera mounted on the bridge of the goggles was aimed forward to record the scene the user was seeing. After calibration, the software overlays a dot corresponding to gaze direction onto the video from the forward-facing camera, and researchers then use the dot to determine where the participant was looking at any point in time. The eye-tracker goggles are shown in figure 61, and screenshots from the eye tracker software showing pupil identification and dot overlaid on a scene are shown in figure 62.

Figure 61. Photo. MPI system performance experiment—eye-tracker goggles.

Figure 62. Screenshot. MPI system performance experiment—pupil location (left) and dot representing gaze direction overlaid on scene (right).

Facilities and Equipment

The experiment was performed on the Smart Road using one of the same test vehicles as in the scoping experiment. The vehicle headlamps were aligned each evening before experiments were conducted.

Participants

Participants were recruited and screened as described in chapter 3. A total of 13 participants performed the experiment. Seven were older, and six were younger. Seven were female, and six were male. Mean and standard deviation of participant age, visual acuity, mesopic visual acuity, low contrast visual acuity, and UFOV are listed in table 17.

Table 17. MPI system performance experiment participant characteristics.

Procedure

One participant completed the testing protocol during each session. Upon arrival, participants followed the procedure outlined in chapter 3 but were also fitted for the eye tracker, which was then calibrated. They next proceeded to the Smart Road, where they drove one practice lap followed by eight experimental laps at a maximum speed of 56 km/h (35 mi/h). During the experiment, the confederate pedestrians positioned themselves at the stations and changed the headlamp filters according to the protocol. The in-vehicle researchers configured the MPI system according to protocol and recorded detection and color-recognition distances.

Data Analysis

After reduction, analysis of variance (ANOVA) tests were performed comparing detection and color-recognition distances across the independent variables. ANOVAs show whether the mean values of multiple variables differ significantly from each other. They do not report which means differ significantly from other means. Therefore, Tukey’s honest significant difference (HSD) tests were also performed to determine which individual variables significantly differed from other individual variables. When Tukey HSD results are shown in charts, bars sharing a letter are not significantly different from each other.

Detection Rates

Detection rate is the percentage of instances a pedestrian, either on the same side of the road illuminated by the MPI system or opposite the side illuminated by the MPI system, was detected. Detection rates were only calculated with respect to the pedestrian-illuminated-by-MPI-system variable.

Mean Fixation Duration

Video clips of participants detecting pedestrians were isolated from the video files using button presses as flags. A custom MATLAB® program allowed data reductionists to calculate the fixation duration, i.e., the length of time the participant gazed at a pedestrian. Mean fixation durations were calculated for the three MPI configurations (off, tracking, and highlighting).

Results

Detection and Color-Recognition Distances

For detection distance, three independent variables resulted in significantly different means: age, headlamp color, and pedestrian clothing color. There was also a significant effect of the interaction between MPI system performance and headlamp color. Detection-distance results are shown in table 18.

For color-recognition distance, two independent variables resulted in significantly different means: age and pedestrian clothing color. There was also a significant effect of the interaction between MPI system configuration and headlamp color. Color-recognition results are shown in table 19.

Table 18. MPI system performance experiment detection distance results per independent variable.

Table 19. MPI system performance experiment color-recognition distance results per independent variable.

Results are discussed in detail in following subsections.

Age

Age significantly affected detection distance because younger participants (M = 70.2 m (230 ft)) detected pedestrians from significantly farther away than older participants (M = 49.2 m (161ft)), p < 0.0001).

Age also significantly affected color-recognition distance because younger participants (M = 57.2 m (188 ft)) recognized clothing color from significantly farther away than older participants (M=42.1 m (138 ft)), p < 0.0001).

Headlamp Color

Headlamp color significantly affected detection distances. When using white/yellow headlamps, participants detected pedestrians from significantly farther away (M = 64.5 m (212 ft)) than when using white/blue headlamps (M = 56.7 m (186 ft), F=5.90, p<0.05) (figure 63). Headlamp color did not significantly affect color-recognition distances. When using white/yellow headlamps, participants recognized the clothing color from a mean of 53.0 m (174 ft) away, and when using white/blue headlamps, participants recognized the clothing color from 48.2 m (158 ft) away. The results are shown in figure 63.

1 m = 3.3 ft

Figure 63. Chart. MPI system performance experiment—pedestrian-detection and color-recognition distances versus headlamp color.

Headlamp Color and Pedestrian in Overhead Lighting

Headlamp color significantly affected detection distances when the pedestrian was in sections of the road with and without overhead lighting. In both cases, detection distances were greater with the white/yellow headlamps than with white/blue headlamps, as illustrated in figure 64.

1 m = 3.3 ft

Figure 64. Chart. MPI system performance experiment—pedestrian-detection and color-recognition distances versus headlamp color and overhead lighting.

The effects of headlamp color on detection and color recognition were robust enough to be replicated in different ambient-light scenarios. One would expect white/blue to have greater detection and color-recognition distances than white/yellow, because mesopic vision is more sensitive to blue. However, pedestrian detection was performed with participants free to move their heads and eyes. Thus, detection probably occurred in the fovea, less sensitive to mesopic effects than peripheral vision. The transmittance of the white/yellow filters was 49 percent, and the white/blue filters was 44 percent, making the white/yellow headlamps slightly brighter, perhaps causing the greater detection distances.

MPI System Configuration

The MPI system was configured three ways: set to off, set to appear to track the pedestrian, or set to appear to highlight the pedestrian.

The ANOVA revealed that the MPI system configuration’s effect on detection distance approached significance at the 0.10 level (F = 2.5, p = 0.1076). A Tukey HSD test found that pedestrians were detected from significantly farther away when the MPI system was off (M = 67.4 m (221 ft)) than when it highlighted (M = 56.7 m (186 ft)) or tracked pedestrians (M = 61.1 m (200 ft)), as shown in figure 65. That could be because the MPI system’s actions distracted the participant, making it more difficult to detect pedestrians.

MPI system configuration did not approach significance with respect to color-recognition distance (F = 0.52, p = 0.6037).

1 m = 3.3 ft

Figure 65. Chart. MPI system performance experiment—pedestrian-detection and color-recognition distance versus MPI system configuration.

MPI System Configuration and Pedestrian in Overhead Lighting

There was no significant effect on detection distance from the interaction between MPI system configuration and whether the pedestrian was in overhead lighting. In the area with no overhead lighting, pedestrian-detection distances were similar to the overhead-lighting condition for the three MPI system configurations. In the area with overhead lighting, the pedestrian-detection distance for the highlighting MPI system configuration (M = 51.4 m (169 ft)) was shorter than for the other two configurations (tracking (M = 61.2 m (201 ft)) and MPI off (M = 65.0 m (213ft)), visible in figure 66, but not to a statistically significant degree.

There was also no significant effect on color-recognition distance from the interaction between MPI system configuration and whether the pedestrian was in overhead lighting.

1 m = 3.3 ft

Figure 66. Chart. MPI system performance experiment—pedestrian-detection and color-recognition distances versus MPI system configuration and overhead lighting.

MPI System Configuration and Headlamp Color

There was a statistically significant interaction between the headlamp color and MPI system configuration (F = 3.90, p < 0.05). When the MPI system was tracking the pedestrians, headlamp color had almost no effect on detection distance, which was about 61.0 m (200 ft). When the MPI system was off, mean detection distance for the white/yellow headlamps (M = 70.5 m (231ft)) was slightly greater than for the white/blue headlamps (M = 63.6 m (209 ft)). The greatest effect was when the MPI system was highlighting the pedestrian. With the white/yellow headlamp, the mean detection distance (M = 63.7 m (209 ft)) was significantly greater than with the white/blue headlamp (M = 49.2 m (161 ft)). The results are shown in figure 67.

1 m = 3.3 ft

Figure 67. Chart. MPI system performance experiment—pedestrian-detection and color-recognition distances versus MPI system configuration and headlamp color.

MPI System Configuration and Pedestrian on Left or Right

There was no statistically significant effect on detection distance from the interaction between MPI system configuration and pedestrian on the left or right of the road. The shorter detection distances for the highlighting MPI system configuration were apparent whether the pedestrian was on the left of right side of the road.

There was no statistically significant effect on color-recognition distance from the interaction between MPI system configuration and pedestrian on the left or right of the road.

The mean detection and color-recognition distance for pedestrians on the left of the road with the MPI system off was visibly greater than the other means (figure 68). This could possibly be because of a combination of the vehicle’s blind spot on the left and the distracting headlamp action. Those two factors could have shortened detection and color-recognition distances when the pedestrian was on the left and the MPI system was on.

1 m = 3.3 ft

Figure 68. Chart. MPI system performance experiment—pedestrian-detection and color-recognition distances versus MPI system configuration and pedestrian on left or right.

Pedestrian Clothing Color

Clothing color significantly affected detection distance. Pedestrians wearing red (M = 66.2 m (217 ft)) were detected significantly farther away than pedestrians wearing blue (M = 55.6 m (182 ft), p = 0.0015), as shown in Figure 69.

Clothing color also significantly affected color-recognition distance. Participants recognized the clothing color of pedestrians wearing red (M = 57.9 m (190 ft)) from significantly farther away than pedestrians wearing blue (M = 43.0 (141 ft), p = 0.0002), also shown in figure 69.

1 m = 3.3 ft

Figure 69. Chart. MPI system performance experiment—pedestrian-detection and color-recognition distances versus pedestrian clothing color.

Pedestrian Clothing Color and Pedestrian in Overhead Lighting

When detection distances were broken down by whether the pedestrian was in an area with overhead lighting, clothing color did not have a significant effect on detection distance for either overhead-lighting condition. With no overhead lighting, red-clothed pedestrians (M = 70.6 m (232 ft)) were detected from farther away than blue-clad pedestrians (M = 56.3 m (185 ft)). With overhead lighting, red-clothed pedestrians (M = 61.4 m (201 ft)) were also detected from farther away than blue-clad pedestrians (M = 54.5 m (179 ft)), but this interaction of presence/absence of overhead lighting and pedestrian clothing color was not found to be statistically significant.

The same trend held for color-recognition distances. With no overhead lighting, participants recognized the color of red-clothed pedestrians (M = 61.3 m (201 ft) from farther away than blue-clad pedestrians (M = 44.6 m (146 ft)). With overhead lighting, participants recognized the color of red-clothed pedestrians (M = 54.1 m (177 ft)) from farther away than blue-clad pedestrians (M = 41.0 m (134 ft)), but again, this was not statistically significant. Both detection and color-recognition distance results by pedestrian clothing color and overhead lighting are shown in figure 70.

1 m = 3.3 ft

Figure 70. Chart. MPI system performance experiment—pedestrian-detection and color-recognition distances versus pedestrian clothing color and overhead lighting.

Pedestrian on Left or Right

Pedestrians on the right side of the road (M = 62.6 m (205 ft)) were detected from farther away than pedestrians on the left side of the road (M = 58.7 m (193 ft)) but not to a statistically significant degree.

Participants recognized the clothing color of pedestrians on the right side of the road (M = 53.2 m (175 ft)) from farther away than clothing color of pedestrians on the left side or the road (M = 47.7 m (157 ft)); this effect was also not statistically significant. Both effects are illustrated in figure 71.

1 m = 3.3 ft

Figure 71. Chart. MPI system performance experiment—pedestrian-detection and color-recognition distances versus pedestrian on left or right.

Pedestrians could be detected and their clothing colors recognized from farther away when they stood on the right-hand side of the road because of the headlamp beam pattern: the headlamp cutoffs that prevent light from creating glare for oncoming vehicles also prevent light from shining on the left-hand shoulder of the road and illuminating pedestrians there.

Pedestrian in Overhead Lighting

The mean detection distance for pedestrians in areas with no overhead lighting (M = 63.1 m (207ft)) was greater than the mean detection distance for pedestrians in areas with overhead lighting (M = 58.2 m (191 ft), p = .2430). An SNK test indicates that while these results are not statistically significant in terms of an analysis of variance, these mean detection distances were meaningfully different from each other, as shown in figure 72. Results of the mean color-recognition distances were similar in terms of significance for with (M = 48.1 m (158 ft)) and without (M = 53.0 m (174 ft)) overhead lighting.

1 m = 3.3 ft

Figure 72. Chart. MPI system performance experiment—pedestrian-detection and color-recognition distances versus pedestrian in overhead lighting.

It should be noted that for this comparison and the clothing color and overhead-lighting conditions, Stations with overhead lighting had more visual clutter than the stations without overhead lighting, and that clutter could have masked the pedestrians, as illustrated in figure 73. Therefore, the above results must be interpreted with caution.

Figure 73. Photo. MPI system performance experiment—visual clutter on the Smart Road.

Detection Rate by Pedestrian Illuminated by MPI System

Some trials were performed with a pedestrian on the side of the road opposite the location where the MPI beam tracked to determine whether the MPI system interrupted the participants’ normal scanning behavior.

When the MPI system illuminated the pedestrian, participants detected the pedestrian about 70 to 80 percent of the time. When the MPI system illuminated the roadway opposite the pedestrian, participants detected the pedestrian about 50 to 60 percent of the time. The results, illustrated in figure 74, show that participant drivers either trusted the MPI system or expected the experimental configuration to be predictable.

Figure 74. Chart. MPI system performance experiment—percentage of detections of pedestrian versus MPI system configuration and whether the pedestrian was illuminated by the MPI system.

Mean Fixation Duration by MPI System Configuration

When the MPI system was configured to track the pedestrian, it began illuminating the pedestrian when the vehicle was 183 m (600 ft) away and swiveled to keep the pedestrian illuminated as the vehicle approached. When the MPI system was configured to highlight the pedestrian, it shined on the pedestrian when the vehicle was 46.7 m (150 ft) away.

Mean fixation duration per MPI system configuration was calculated, but the fixation duration data was not normally distributed, so comparing means to determine statistical significance was not valid. Therefore, a non-parametric statistical analysis, the Kruskal-Wallis test, was used to compare the groups of fixation durations between MPI system configurations. There was no statistically significant difference in fixation duration between the three MPI system configurations (p = 0.8040). The mean fixation durations by MPI system configuration are shown in figure 75.

Figure 75. Chart. MPI system performance experiment—mean fixation duration versus MPI system configuration.

Discussion

An objective of this project was to evaluate the effect of a peripheral illumination system on driver visibility. In the absence of overhead lighting, headlamps were the only source of light the driver had available to aid in object detection, and the effects of the MPI system were isolated. Results were mixed, with the MPI system increasing participant detection rates for pedestrians but decreasing detection distances. This result indicates that in a vehicle with an active MPI system, drivers rely on the MPI system to help them detect pedestrians. An MPI system that can detect and highlight pedestrians in the periphery better than human scanning alone would likely increase detection rates for those pedestrians. However, the MPI system may distract the driver’s attention from the roadway, because the driver would follow the moving headlamp beam. So while the system may distract in one case, it focuses the driver’s attention in another. Refocusing a driver’s attention may not always lead to better driving, as seen in the cases where the MPI headlamp beam was directed to a position opposite the pedestrian, reducing detection distances and rates for those pedestrians.

In instances where pedestrians were located on the right side of the vehicle, the MPI highlighting and tracking configurations resulted in detection distances similar to when the MPI system was inactive. However, when pedestrians were located on the left side of the vehicle, the inactive MPI scenarios resulted in the longest detection distances. A possible explanation for this would be the design of the vehicle, where the A pillar may have impeded the driver’s view of the left-hand side of the roadway. With a moving headlamp, drivers would have a higher cognitive load trying to search for pedestrians around the A pillar than trying to search for them on the right side of the road with no obstruction,.

A significant interaction took place between headlamp color and the behavior of the MPI system, resulting in findings that are more telling than those found in the scoping experiment, which had too few off-axis detections to draw firm conclusions. The findings from this experiment showed that the white/yellow headlamps might have allowed drivers to detect peripheral pedestrians from farther away, possibly indicating an effect of the light source’s SPD.

Conclusions

The first objective of this experiment was to evaluate how MPI system configuration affected driver visual performance. Results showed that participants detected pedestrians from the greatest distances when the MPI system was off. That was the case for sections of the road both with and without overhead lighting. This could be because the MPI system’s moving beam distracted the participants and prevented them from scanning normally. Additionally, participants seemed to assume the MPI system would accurately illuminate pedestrians; they detected the pedestrian between about 70 to 80 percent of the time with the MPI system illuminating the pedestrian, compared with about 50 to 60 percent of the time with the MPI system illuminating the roadway opposite the pedestrian. This could have serious implications for an MPI system that failed to detect all pedestrians or malfunctioned to illuminate a “false-positive” pedestrian; drivers might assume the system was working and be less vigilant, causing more pedestrian crashes and casualties.

Other research objectives of the experiment was to evaluate how headlamp color of the MPI system affected driver visual performance and how the MPI system and overhead lighting interacted to affect driver visual performance. The white/yellow headlamps had greater detection distances than white/blue headlamps in sections of the road with and without overhead lighting. That could be partially because the white/yellow gels had slightly higher transmittance than the white/blue gels (see table 3). Although mesopic vision is more efficient for blue than for yellow light, drivers were allowed to turn their heads and move their eyes as they scanned the road, and pedestrian detection occurred foveally where mesopic effects are minimal.

An eye tracker was used to determine how the MPI system affected driver eye-glance behavior, the last research objective of the experiment. There were no statistically significant differences among the different MPI system configurations for glance duration.

Future Research

The experiment described in this chapter was performed with a single overhead-lighting color with pedestrians placed at the same distance from the shoulder for each run. To better characterize mesopic vision and its interaction with an MPI system, future experiments should use other overhead-lighting colors and add off-axis angles for object detection. The final experiment in this project takes these concerns into account.

Drivers typically glance around the roadway as they drive. Future analyses should characterize exactly how an MPI system affects glance behavior.

Participants apparently trusted the MPI system to the extent that they missed important visual information. Therefore, any MPI system would have to be accurate enough that it would seldom, if ever, miss detecting a pedestrian on the side of the road.