Traffic Control Device Conspicuity

MDS Classification of Roadways

If perceptions of the roadway environment are important to the behavioral response to TCDs, then methods of identifying determinants of those perceptions are likewise important. There have been many attempts to accomplish this task and classify our visual environment in a meaningful way.^(11,12) The Federal Highway Administration (FHWA) employs a functional-use system with the following classifications: interstates, other freeways, principal arterials, minor arterials, major collectors, minor collectors, and local roads.⁽¹³⁾ This system relies on road functionality but ignores other environmental characteristics. For example, arterial roads in urban and rural areas can look vastly different. An arterial in a suburban or rural area may consist of two travel lanes and be surrounded by foliage and minimal signage. An arterial in an urban environment may contain six travel lanes and be surrounded by buildings and ample commercial signage, or with sufficient screening foliage, an urban arterial may look very much like a two-lane rural highway. Functional classes that do not reliably predict roadway appearance may not be a problem if the salient characteristics of roadway class determine driver responses to TCDs. However, alternative classification systems exist, and it is not known which classification systems best predict when drivers will notice and respond to TCDs.

Alternative roadway classification systems may rely on situational components-items that move freely to and from the area (e.g., pedestrians, vehicles). Items in a scene can also be classified as built, permanent components (e.g., grocery stores, trees). Whereas these types of functional use and component classification systems are quite useful in some respects (e.g., the implementation of speed limits, curbing, and sidewalks), they are unable to provide substantial objective information about environmental influences that may be vital to driver performance (e.g., detection of speed limit signs, stop signs, and construction zone warnings). Furthermore, these classifications fail to capture many driver-relevant perceptual differences between environments.

When performing a visual search task (looking for a specific target item among other items), the more non-target items present, the longer it takes to complete the task.^(14-16) However, the exact manner for counting items is debatable. For instance, a person could be counted as a single item or as multiple items if shirt, shoes, pants, nose, fingers, etc. are considered separate elements. Further, the number of non-target items (set size) alone does not dictate the amount of time required to look at and identify a specific target object. In many cases the target item "pops out" (e.g., a red line among black lines).⁽¹⁷⁾ People also have a tendency to direct attention toward items of relevance or guide attention to specific task-relevant areas of a visual environment.^(18,19) For instance, drivers in countries that drive on the right look for signs on the right side of the road and may miss the same signs when they are placed on the left side of the road.⁽²⁰⁾

Drivers tend to look at task-relevant signage. However, in visually complex environments, excessive clutter may increase the time required to identify, interpret, and respond to task-specific stimuli. This delay may lead to unsafe driving behavior.⁽¹⁵⁾ It is easy to imagine that a driver might miss a relevant sign (e.g., lane ending) among a plethora of other signs and heavy traffic. As a result, the driver may be startled by a sudden change in the roadway or behave erratically.

It is obvious that the roadway environment is complex and it can be difficult to determine exactly what visual information degrades driver performance and what does not. It would be beneficial to increase the saliency of environmental components that improve driving performance. A first step toward accomplishing that goal is to better understand which items drivers attend to in a roadway environment. Such information could then be exploited in roadway design by directing drivers' attention to pertinent items and away from the non-essential. This study sought to increase the understanding of drivers' perceptions of the roadway environment and to use this understanding in the interpretation of data from subsequent studies that examined driver eye glance behavior and detection and identification of highway signs. Participants were asked to rate the similarity of various photographs of roadway environments. The analysis of observer perceptions of roadway environments utilized MDS, a somewhat novel methodology for the realm of transportation research.⁽²¹⁾

At the simplest level, MDS is a mathematical technique that recovers the spatial relationships between objects based on the distances between those objects. For example, Kruskal and Wish demonstrated that a map of the location of 10 major U.S. cities could be accurately reconstructed by submitting the distances between all pairs of those cities to an MDS.⁽²²⁾ Other methods exist for accomplishing this task when all paired distances are known and measured without error. MDS becomes useful when measurements cannot be made without error. That is, whereas the spatial distances between cities can be measured with great accuracy, subjective judgments of the similarity of objects, such as pictures of outdoor scenes, may reflect both inter- and intra-individual variability. Furthermore, if it is assumed that people perceive the similarity of objects based on multiple factors and that those factors can be conceived of in spatial relationships, then MDS can be used to tease out the factors that define the space. For instance, Wish asked students to make similarity judgments between pairs of nations constructed from a list of 12 nations. He identified two perceptual dimensions that predicted the spatial relationships extracted from MDS analysis: degree of industrialization (development) and communist/non-communist political orientation.⁽²³⁾ The latter dimension probably reflected the Cold War era during which Wish conducted his study.

MDS analyses might provide valuable clues as to the how drivers perceive roadway environments. When combined with the studies that follow, this type of information has the potential to extend understanding of the interactions between roadway sign placement, lane markings, and roadway geometry with other environmental factors that influence driver performance.

Method

Participants were asked to rate the similarity of roadway scenes. These similarity ratings were then used to infer a mental model of the scenes.

Stimuli

The roadway scenes were panoramic photographs of the environment surrounding the TCDs that served as the stimuli of interest in subsequent studies. The environments around 21 TCDs were photographed from a point 85 ft (26 m) upstream of the target TCD so as to present a driver's view as the TCD was approached. Each panorama consisted of three photographs that were stitched together to form a wide-angle panorama. All three photographs were shot from the right through lane of the roadway. The center photograph approximated a driver's view of the road and centered on the right lane at a point slightly beyond the target TCD. The left and right photographs bracketed the center photograph with approximately 30 percent overlap. From the original set of 21 panoramas, 14 were chosen for the similarity comparisons. The down selection was intended to maintain the widest range of roadway environments while limiting the number of paired comparisons needed.

Procedure

Participants were asked to rate the similarity of 91 unique pairs of the 14 panoramic roadway photographs (see figure 1 through figure 14). They rated the pairs on a scale from 1 (not at all similar) to 10 (very similar). Each panorama was presented on a laptop computer at approximately 960 (width) by 304 (height) pixels. Each panorama was presented as the upper or lower member of a pair an equal number of times.

Figure 1. Photo. Panorama 1.

Figure 2. Photo. Panorama 2.

Figure 3. Photo. Panorama 3.

Figure 4. Photo. Panorama 4.

Figure 5. Photo. Panorama 5.

Figure 6. Photo. Panorama 6.

Figure 7. Photo. Panorama 7.

Figure 8. Photo. Panorama 8.

Figure 9. Photo. Panorama 9.

Figure 10. Photo. Panorama 10.

Figure 11. Photo. Panorama 11.

Figure 12. Photo. Panorama 12.

Figure 13. Photo. Panorama 13.

Figure 14. Photo. Panorama 14.

After rating the similarity of each of the pairs of photographs, participants were asked to rate each of the roadway photographs individually on five different descriptors: (1) built-up, (2) clutter, (3) openness, (4) aesthetically pleasing, and (5) organized/predictable. These ratings were on a 1-10 scale with 1 representing the high end of the scale and 10 representing the low end of the scale (e.g., 1 for "very built-up" and 10 for "not at all built-up"). These descriptors were derived from pilot testing in which colleagues were asked to generate a verbal description of the panoramic photographs. Participants were not provided definitions or examples for the descriptors but were left to interpret the terms on their own.

Thirteen people (7 males, 6 females) provided similarity ratings. All participants were recruited at a local community center, where they also completed the descriptor ratings. The mean age of participants was 39.8 years (range 24-63). All participants had a valid driver's license. MDS can be performed on similarity rating from as few as one individual. Unless individual differences are of interest, sample sizes in the range used for this study are generally adequate.

Results

Table 1 shows the x, y coordinates of the MDS results for a two-dimensional solution.

Table 1 . MDS coordinates of a two-dimensional solution for the panorama similarity ratings.

The MDS routine PROXSCAL was used to analyze the similarity ratings.⁽²⁴⁾ The data were treated as interval scale. ⁽²⁴⁾ An examination of the scree plot suggested a two-dimensional solution (stress = 0.117). The two-dimensional MDS solution is shown in figure 15. The number beside each point in the plot refers to the panorama ID number. The labels on the axes are not MDS outputs; they have been supplied by the authors.

Figure 15 . Graph. MDS solution for 14 panoramas along field study route.

The Pearson product-moment correlation between each of the five descriptor ratings of the panoramas and the MDS dimension scores of the panoramas was computed. As shown in table 2, each of the MDS dimensions correlated significantly with ratings of at least one of the preselected descriptors. The first MDS dimension significantly correlated with three descriptors: built-up (r = 0.56, p < 0.05), clutter (r = 0.75, p < 0.01), and aesthetically pleasing (r = ‑0.66, p < 0.01). The descriptor ratings built-up and clutter strongly positively correlated with one another (r = 0.78, p < 0.01). Thus, those environments with a high degree of manmade or built components were also judged to be cluttered. The first MDS dimension was also negatively correlated with aesthetically pleasing. Aesthetically pleasing was not significantly related to any of the other descriptors. Thus, it appears that dimension 1 may reflect the participants' perception of clutter; however, the possibility that the correlation is coincident with some factor that was not explored cannot be ruled out-MDS is an exploratory technique. If a scene is judged to be high on clutter, it is generally judged to be low with respect to being aesthetically pleasing. The inverse is also true.

The second dimension most closely correlated to the predictability or organization of the environment (r = -0.55, p < 0.05). Items that had high scores on the second MDS coordinate were judged to have low predictability or organization. Ratings of organization and predictability had a significant positive correlation with aesthetically pleasing (r = 0.826, p < 0.01).

Table 2 . Matrix of correlations between each descriptor rating and each MDS dimension.

*Significant at 0.05.
**Significant at 0.01.

Discussion

This study used MDS to identify environmental factors that observers attended to when comparing roadway scenes. Two dimensions emerged that mapped well to previously suggested roadway environment descriptors. These factors provide insight into the types of characteristics that are salient to roadway users. It appears that participants attended to environmental components that have the potential to affect driving behavior. For example, an organized/predictable environment may allow drivers to better predict where vehicles are likely to turn or pedestrians are likely to cross.

The study utilized a range of roadway scenes that were present along the intended route for a field study of driver behavior. Because of constraints on the selection of TCDs to be included in that field study, the range of variability in the pictured environments may have limited the range of environmental factors that emerged from paired comparison ratings. It is important that subsequent research in this area include a wider variety of roadway environmental factors, such as traffic, nighttime scenes, greater variability in access control, a greater range of roadside advertising, and a greater range of pedestrian and bicycle activity. MDS relies on comparisons of nearly exhaustive pairings between items. As a result, the number of items compared generally needs to be limited to around 15 to avoid overwhelming participants. Nonetheless, MDS may prove to be a robust methodology for identifying aspects of an environment attended to by drivers. As a result, use of MDS in combination with other research tools has the potential to set the foundation for a driver-centric roadway classification system. For signing and marking applications, such an approach may be as or more useful in guiding practitioners than functional classification systems. Future studies may explore other methods of quantifying visual scenes, such as automated image analysis. Such methods combined with driver-centric techniques have the potential to greatly expand our understanding of the interaction between users and roadway environmental factors.

Each of the 14 panoramas used in the MDS analysis included a TCD that was a target for recall in the field study that followed. Each panoramic photograph was taken 85 ft (26 m) upstream of a TCD for which recall was requested after the driver passed the TCD. One question to be addressed in the analysis of the field study data is whether the MDS classifications predict either glances to or recall of the TCDs.