U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
|This report is an archived publication and may contain dated technical, contact, and link information|
|Publication Number: FHWA-HRT-13-044 Date: August 2013|
Publication Number: FHWA-HRT-13-044
Date: August 2013
In the glance behavior and sign recall study, drivers were asked to identify speed limit and warning signs about 2 s after the signs were passed. It was found that 24.3 percent of the time drivers could not identify signs that they had just gazed at but correctly identified 53.3 percent of the signs they had not looked at directly. It is possible that some signs that had not received looks were guessed from the roadway context. However, the selection of signs queried was biased toward signs for which contextual cues would be minimal. For instance, the most common warning sign, the pedestrian crossing sign, was not included, and non-standard warning signs (e.g., parked vehicle ahead) were included. Speed limit signs were only queried after a change in speed limit or on a new road, so recall of previous speed limit signs was probably not a factor in speed limit recall. Reduced speed ahead signs did not precede any of the targeted speed limit signs. In most cases, there were no marked changes in roadway environment in proximity to the speed limit signs.
Only the central 2° of vision, foveal vision, provide resolution sharp enough for reading or recognizing fine detail.(41) Useful information for reading can be extracted from parafoveal vision, which encompasses the central 10° of vision.(41) Rayner et al. suggest that foveal vision encompasses 6-8 characters of normal-size printed text.(41) They presented sentences to participants and used an eye tracker to move a mask and hide the portion of the sentence around the participant's fixation point. When the mask covered the width of 17 characters of text, participants were still able to read text at the edges of the mask, albeit at a rate of only about 10 words per minute. The text not hidden by the mask was in the participants' peripheral vision. Thus, drivers might be able to read traffic signs, which generally have large symbols or letters, without fixing their gaze on them. As with sign detection, the surrounding environment may affect the ability to read the sign. The objective of the present study was to measure off-axis sign-reading ability in the context of various surrounds. The fixation point offset from the signs was varied from 0 to 15° in the context of five backgrounds used in the MDS classification study.
In a Finnish study, peripheral perception of traffic signs was examined for offsets from a fixation point of 10-80°.(51) Perception of sign color and shape was examined in addition to sign identification. Color and shape perception was accurate to about 50°. Sign identification was relatively accurate at 20°, but most of the signs used were easily distinguished from each other based on shape or color differences. Discrimination at 20° was essentially chance when the identification relied on reading text or distinguishing between symbols on otherwise identical signs. The study did not assess whether drivers could distinguish speed limits or text warnings in parafoveal or peripheral vision.
For signs that subtended 4° of visual angle, Karttunen and Hakkinen found no effect of background scene on peripheral sign perception.(51) However, the authors did find that the inclusion of signs above and below the target sign diminished correct perception of the target sign. This suggests that similar signs located close to a target reduce identifiability. Such an effect is known as crowding.(52) Crowding was not the focus of the present study, which instead focused on the effect of the broader environment in which signs are located rather that the local effects of clutter in the immediate surrounds.
In the experiment reported here, signs embedded in various roadway backgrounds were projected onto a screen. The visual angle between the signs and the participant's fixation point was varied. The dependent measure was identification of the signs' messages. Speed limit and text-based warning signs were used. Given the results of the glance behavior and sign recall study, it was hypothesized that both types of messages would be readable at substantial off-axis angles. The findings of Rayner et al. suggested that sign reading may occur with glance offset up to 10°.(41)
The same driving simulator was used in the same manner as in the sign detection conspicuity study.
The following TCDs were used:
Each TCD was presented in each of six roadway contexts. These contexts or scenes served as an independent variable. The scenes were the six panoramas shown in figure 1 through figure 6. Each scene was composed of three photographs stitched together to form a panorama of about 120°. The center photograph in the panorama was always aligned with the roadway and centered on a point in the right through lane somewhat beyond the TCD location. In some locations, there was an acceleration or deceleration lane to the right of the through lane.
Sign stimuli used in this experiment were presented within the panoramic highway scenes at locations on the right side of the road, where such signs might be expected and where an actual TCD 85 ft (26 m) from of the camera had been present in the original photographs.
Speed limit and warning sign identification was tested in separate blocks of 240 trials. The order of blocks was counterbalanced across participants. Each trial began with the presentation of a fixation cross that subtended 2° of visual angle on a gray background. The fixation cross was displayed for 1 s. The cross was followed by a panoramic scene that was displayed for 0.15 s. The panoramic scene was followed by a gray background.
The fixation cross was vertically aligned with the target sign in the subsequent scene. The horizontal offset of the fixation cross from the location of the target TCD was varied randomly from trial to trial. Each participant received a different random order of scenes, offsets, and signs. Offsets of -9, -6, 0, 3, 6, 9, 12, and 15° were used. The target signs were all on the right side of the road, and the photographs were centered on the roadway at a point beyond the signs. Scene factors such as lane width and horizontal and vertical curvature resulted in shifts in screen location of signs as a function of scene. Figure 29 provides an example of where the fixation cross was positioned relative to a target sign. The horizontal line illustrates the vertical positioning of the fixation cross. Five of the eight offset angles are labeled in the figure.
Participants were provided short breaks every 80 trials. Five practice trials preceded each block of 240 trials, one practice trial with each sign. The practice trials all had a 0° offset. The background scene for the practice trials was not one of the six scenes used on test trials but was created in the same manner as those scenes.
Twelve individuals were tested. Nine were male, and three were female. The mean participant age was 35 years (range 27-50). All participants were licensed drivers and had 20/30 or better foveal visual acuity (with correction if necessary) in each eye.
Participants were instructed, "We are interested in how far a sign can be from the point of gaze and still be read." They were asked to turn their head and gaze toward the fixation cross and to keep their gaze at the cross's location even when the roadway scene appeared. Participants were urged, "Remember, always keep your eyes fixed on the location of the cross. Do not shift your gaze to the sign." Participants were informed of the type of signs to be presented in the upcoming block (i.e., speed limit or warning signs) and shown an array of pictures of the five signs in that block. They were asked to call out the speed limit or text on the sign as soon as possible after the sign appeared.
The proportion of correct sign identifications was analyzed as a function of sign type (speed limits or warnings), background scene (six scenes), and degree of offset of the fixation point from the signs (eight offsets). A 2 (sign type) by 6 (scene) by 8 (offset) repeated measures analysis of variance was used. The proportion correct for speed limit signs was computed by averaging 1 (correct) and 0 (error) values over the five speed limit signs. Likewise, the proportion correct for warning signs was computed by averaging 1 and 0 values for the five warnings. Thus, each participant provided 48 data points. An arcsine transformation was performed on the proportion correct before performing the analysis of variance.(53) This transformation resulted in a distribution of scores that better met the analysis of variance normality assumption. Partial eta squared values were calculated to indicate effect size.
As shown in figure 30, speed limit identification performance was superior to warning sign identification performance, F(1,11) = 221.2, p < 0.001, n2p= 0.95. Identification performance for both sign types decreased as the angle away from the fixation point increased, F(1,11) = 35.8, p < 0.001,n2p= 0.77.
There was an interaction between sign type and scene background, F(5,55) = 2.9, p = 0.04, n2p= 0.21, that resulted because background scene had a significant effect on warning sign identification, F(5,55) = 2.9, p = 0.03, n2p= 0.24, but not on speed limit sign identification (p = 0.11). These effects are shown in figure 31, where error bars represent standard errors. Post‑hoc paired comparisons showed that warning sign identification with scene 3 was significantly better than with scenes 2 and 4 (p < 0.01) and that performance with scene 5 was marginally superior to that in scene 4 (p = 0.04). No other paired comparisons yielded statistically significant performance differences.
With five signs to select from, chance sign identification performance was 20 percent. Even with a 15° offset, average performance remained above chance. This confirms that traffic sign messages can be recognized in the absence of a fixation. Thus, failures to look at signs cannot be taken as evidence that drivers are unaware of signs. Awareness of sign content, then, is a better measure of sign conspicuity than fixation on a sign. This does not mean that glance data are not useful. When glances do not fall within 9° of a sign, the probability of understanding the sign is markedly reduced. This effect is likely to be more pronounced in an actual roadway environment than in a laboratory environment where the participant's only task is to identify signs from a small known set of alternatives.
The difference in performance between speed limit and warning signs is likely the result of the size of the characters. The speed limit numbers were scaled to be equivalent to 12 inches (31 cm) high, whereas the warning sign letters were scaled to be equivalent to 4.5 inches (11 cm) high. The speed limit numbers were scaled to be typical of those on posted on arterial roadways. The warning sign letters were somewhat smaller than typical because the warning diamonds were mistakenly scaled to be 36 inches (91 cm) on the diagonal rather than on the edge. Therefore, the results probably underestimate the off-axis readability of text-based warning signs.
Although some roadway scene environments resulted in significant differences in warning sign identification, these differences were not large. The correlation between sign identification and the background scene dimensions identified in the MDS analysis was examined. No relationship was apparent between sign identification performance and the MDS dimensionality of the scenes. The best warning sign identification occurred with scene 3, which was low in both clutter and predictability. The worst warning sign identification performance was with scene 2, which was lowest on the clutter dimension and neutral on the predictability dimension. Low warning sign identification performance also occurred on scene 4, which was highest on the predictability dimension and neutral with respect to clutter. Scene 5, which yielded slightly better warning sign identification performance than scene 4, was also neutral with respect to clutter but was at the low end of the predictability dimension. Thus, within the range of scene environments explored in this study, environment had little obvious systematic influence on the ability to distinguish between warnings or speed limits.
The failure to find an interpretable effect of background scene on sign identification performance replicates the findings of Karttunen and Hakkinen, whose background scenes varied over a greater range of environments than those in the present study.(51) However, the traffic signs used in the Karttunen and Hakkinen study were larger and differed more from each other. Schnell, Aktan, and Li also found a lack of background effect on sign identification, but in their study, the sign always appeared in the same location and there was no attempt to limit the signs to peripheral vision.(54) Although background clutter would seem, intuitively, to affect perception of a sign's message, such an influence is difficult to demonstrate. Karttunen and Hakkinen showed a decrement in sign identification performance when they juxtaposed other traffic signs with the target signs, so there is evidence to show that background matters in sign identification.(51) However, there is insufficient research to provide guidance on how to enhance or prevent degradation to sign identification performance.
It is possible that some participants shifted their gaze to the signs on some trials despite the instruction to keep their gaze at the fixation cross location. Indeed, this was likely the case for two participants who missed very few identifications and whose performance did not appear to vary with offset. However, even when the data of these two participants are excluded, mean performance at the 15° offset remained above the chance level of 0.2, and the performance curves remained similar in shape to those in figure 30. Limiting sign and scene exposure to 0.15 s was intended to reduce the ability of participants to shift their gaze from the fixation cross location to the anticipated sign location. However, gaze can be shifted by a saccade in about 0.02 s, and useful information can begin to be extracted about 0.02 s after a saccade has been completed.(41) Thus, participants who shifted their attention when the scenes were first displayed, either knowingly or unknowingly, could have had up to 0.11 s to fixate and read the signs. An eye-tracking device would be necessary to rule out such shifts in fixation. Nonetheless, the drop-off in performance as a function of offset and the fairly symmetrical drop-off between equal negative and positive offsets suggest that most participants were compliant on most trails and that the present data are interpretable even if they may somewhat overestimate legibility as a function of gaze offset.
The signs in this study were located at or near the edge of the pictured travelled way, which is typical of urban signing practice.(35) To provide a measure of safety to run-off-road vehicles, the MUTCD recommends a 12-ft (3.7-m) lateral offset from the edge of the travelled way or an 8-ft (2.4-m) offset from the edge of an 8-ft (2.4-m) shoulder.(36) From a distance of 85 ft (26 m), a 12‑ft (3.7-m) offset would put the sign about 15° to the right of the driver's forward gaze. From longer distances, the angle would be less. The present results suggest that the likelihood that signs will be read in the absence of direct fixations on them would increase if offsets less than 12 ft (3.7 m) were used or if warning signs were larger than current recommendations, which are 36 inches (110 cm) for conventional roads and 48 inches (146 cm) for freeways.