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-05-080
Date: May 2006
Pedestrian Access to Roundabouts: Assessment of Motorists' Yielding to Visually Impaired Pedestrians and Potential Treatments to Improve Access
PDF Version (1.14 MB)
PDF files can be viewed with the Acrobat® Reader®
CHAPTER 2. STUDY 1: CLOSED-COURSE EVALUATION
There were two experimental conditions in this study: a control condition and a treatment condition. In the treatment condition, rumble strip-like devices were placed on the pavement surface. Seven severely visually impaired individuals participated. These individuals regularly take pedestrian trips without sighted companions and at least occasionally make such trips to locations that they have not visited before.
The basic requirements for a sound cue were: (1) vehicles approaching the crosswalk would trigger it, (2) the treatment would be relatively inexpensive to implement, and (3) the sound cue would provide distinctly audible cues in a roundabout environment. Initially, commercially available rumble strips, traffic counter tubes, and garden hoses were considered. The first challenge encountered was that cars about to yield at a crosswalk might be traveling at very low speeds. The rumble strips, tubes, and hoses that were evaluated failed to make sufficient noise when passed over at 3 to 11 km/h (2 to 7 mi/h). A treatment was eventually derived that appeared promising for testing, but it would need additional development to serve as a permanent engineering solution. That solution was to cut 3.8 cm (1.5 inch)-diameter polyvinylchloride (PVC) pipe lengthwise into thirds, then secure the pipe, concave side down, to the roadway with asphalt tape. Asphalt tape is normally used to secure traffic counting tubes. To prevent heavy vehicles from crushing the pipe, pieces of dowel cut lengthwise were placed between the PVC pipe and the roadway to provide additional support. Even when the approaching vehicle was driven at 3 km/h (2 mi/h), the PVC pipe device produced a clacking sound that could be heard over ambient traffic noise from a distance of about 20 m (66 ft).
To provide a meaningful sound pattern that visually impaired pedestrians potentially could use, three rows of sound strips were laid transversely across the roadway. One row was placed on the upstream edge of the crosswalk, a second row was placed 6 m (20 ft) upstream of the first row, and a third row was placed 7.3 m (24 ft) upstream of the first row. A two-axle passenger car that passes over the strips generates a rapid "clack, clack, clack, clack" sound as the two axles pass over the two upstream rows, and then, after a brief delay, another "clack, clack" as the vehicle enters the crosswalk and departs. If the first four clacks are produced, but not the latter two, a stopped vehicle is signified. A pair of clacks that are not preceded by four closely spaced clacks indicates that a previously stopped vehicle has departed.
The test was conducted on a rounded, rectangular roadway at FHWA's Turner-Fairbank Highway Research Center in McLean, VA. A scale drawing of the course is provided in figure 4. Although the circuit was somewhat larger than that of a double-lane roundabout, and the 2.9-m (9.5-ft) lanes were somewhat narrow, the course provided a method for approximating traffic movement at a double-lane roundabout. Because the sound-strip treatment could not be easily put down and removed, with- and without-strips test conditions were located at different points on the course, separated by approximately 50 m (165 ft). Platforms were placed 61 cm (2 ft) back from the outer edge of the roadway, adjacent to the two simulated crosswalks. These platforms, intended to simulate a curb, were 20 cm (8 inches) high.
Because only four cars were available to simulate traffic, it was believed that the ambient noise level on the closed course did not adequately represent the noise level at actual roundabouts. Therefore, during testing two loudspeakers were used to broadcast recorded traffic sounds. These loudspeakers were placed approximately 3 m (10 ft) from opposing sides of the pedestrian platforms, such that the A-weighted sound level was 68 decibels at the site where participants stood. This sound level approximated that measured during peak traffic at the roundabout used in study 2.
Seven individuals with severe visual impairments were recruited (four reported having no vision, one reported minimal central vision only, one reported minimal peripheral vision only, and one had some perception of light). Although none of the participants who reported having some residual vision reported using visual cues when traveling, the participant with residual central vision said that he could sometimes detect crosswalk edges if there is high contrast between the pavement and the painted line. All participants who reported having some vision wore a blindfold during the tests. All participants received a hearing screening. All had hearing within the normal range except for one older male (participant 2), who had a low frequency hearing loss in his right ear. Further participant demographic information is provided with the individual debriefing results. Participants were paid $35 per hour, and they completed participation in about 3 hours.
Before data collection began, the participants were briefed and provided an opportunity to explore the mock intersection. The briefing did not include mention of the sound strips, and no participant asked about them during the test session. An orientation and mobility specialist was always present during the pretest, test, and debriefings.
A series of 36 test trials followed 4 practice trials. Scripts governed the behavior of the four vehicles in each trial. The top speed was 24 km/h (15 mi/h), and the speed on turns was lower. The scripts comprised 18 trials that were repeated at each of the 2 crosswalks. Three participants (numbers 2, 4, and 7) were tested in the control condition first, followed by the treatment condition. The remaining four participants were tested in the treatment condition first. The drivers, who were paid for their participation, provided the four vehicles. The same four vehicles were used for all trials of individual participants; each participant performed the control and treatment trials with the same four vehicles; however, different participants worked with different sets of four vehicles. Most of the vehicles were late-model midsize or compact cars. One full-size sport utility vehicle and two full-size pickup trucks were also used; however, not more than one of these was used with any particular participant.
Of the 18 trials, 8 trials called for a vehicle in the near (right) lane to yield first, 6 called for a vehicle in the far (left) lane to yield first, and 4 trials called for vehicles in each lane to yield at the same time.
Throughout the remainder of this document, the near and far lanes are referred to rather than right and left lanes respectively because our focus is from the viewpoint of the pedestrian. For this study, the relationship of the vehicles to the pedestrian is more important than the nominal lane designations from a driver's perspective. In both studies, the pedestrian stood on the sidewalk side of an exit lane. That is, as the pedestrian faced the road while standing on the sidewalk, a vehicle approached from the participant's left side. If the participants had stood on the splitter island side of the exit lane (so that the left lane was the near lane), it is assumed that the near-far relationship would hold with respect to the participant's performance; however, in that case, the vehicles would have approached the crosswalk from the participant's right side.
The scripts roughly balanced the lanes that each of the four vehicles used, and the order in which each vehicle passed the crosswalk. In any given trial, a specific vehicle made between zero and four complete laps. In some trials, one or two of the vehicles did not move. The scripts were designed to obscure the number of vehicles, the amount of time between the start of a trial and its completion, and the time between the first vehicle yield and the second yield. In two trials, the first vehicle to yield pulled away after 10 s, so that in those two trials, both crosswalk approach lanes were never blocked at the same time. The scripts also directed vehicle speed (normal or slow) as vehicles passed the crosswalk. A different randomized order of trials was used for each participant and at each crosswalk.
Figure 5 is an example of the script of one driver for one trial. In the example, vehicle 3 was assigned to the near (right) lane and yielded the first time it arrived at the crosswalk. Vehicle 1, which began at station 'A' upstream (behind) vehicle 3, stopped behind vehicle 3. (The station letters in figure 5 designate the locations where the vehicles waited for trials to begin. These letters correspond to the locations marked by those letters in figure 4.) Vehicle 4 started at station D, which was the closest starting location to the crosswalk, and passed the crosswalk twice (the second time slowly). Vehicle 2 passed the crosswalk once, and then yielded on its second approach to the crosswalk. After the trial was over, the vehicles returned to different stations, so that from trial to trial, a given vehicle would travel in different lanes and a different order with respect to the other vehicles. Each vehicle remained in the same lane throughout a trial and did not pass other vehicles, except when those vehicles stopped at the crosswalk and the trailing vehicle was to continue on another lap.
During each trial, a participant stood on the platform and used hand signals to indicate when he or she detected vehicles stopping or departing after a stop. The participant's signals and the vehicle movements were video-recorded for later scoring.
The primary observed variables were accuracy in detecting blocked lanes (stopped vehicles) and wait time to detect stopped vehicles. Accuracy in detecting when stopped vehicles departed while the other lane remained unblocked was also recorded. Accuracy measures included:
Participants were instructed to use hand signals each time a vehicle stopped at the crosswalk and indicate which lane it occupied. To signal detection of vehicles stopped in the far lane, participants were to point at the vehicle. To signal detection of vehicles stopped in the near lane, participants were to point to their own right. To indicate that two vehicles were stopped side-by-side (i.e., the crossing was blocked), they were to hold a hand over their head. The primary measure of effectiveness for this experiment was accuracy in the detection of both lanes being blocked. This measure was chosen because it was assumed that, with both lanes blocked by stopped vehicles, it might be judged appropriate for a visually impaired pedestrian to make a crossing. With only one lane blocked, a crossing would be less appropriate, because the presence of one idling vehicle creates noise that can mask the approach of other vehicles that might not stop. In the experiment, the participants did not actually make a crossing. The trials ended when the participant indicated that two vehicles were stopped (regardless of whether that indication was correct), or when the participant failed to indicate two stopped vehicles within 10 s of both lanes being blocked. In two trials, the first vehicle to yield waited for 10 s and then drove away before a second vehicle yielded in the other lane. Those trials ended when the participant indicated the second vehicle had stopped, or when the participant failed to indicate that vehicle was stopped within 10 s.
Table 1 shows the results for the detection of double yields (i.e., vehicles blocking each of the lanes). A hit was scored if the participant indicated that both lanes were blocked within 10 s of the second of two vehicles' stopping. A false alarm was scored if at any time before both lanes were blocked, the participant indicated that both lanes were blocked and did not retract that decision within 1 s. A miss was scored if the participant did not detect that both lanes were blocked within 10 s.
As can be seen in table 1, five of the seven participants had superior hit rates with the sound strips in place compared to the results without the strips. This difference approaches statistical significance (z = -2.0, p < 0.05 by the Wilcoxon signed ranks test).(17) However, every participant had at least one false alarm in the treatment condition. False alarms are cause for concern because they imply that the individual might make a crossing in the belief that both lanes are blocked by stopped vehicles when, in fact, they are vulnerable to vehicles approaching at speed. False alarm rates with and without the sound strips were roughly the same. The slightly superior hit performance in the treatment condition came from a corresponding reduction in misses. In the two trials in which both lanes were never blocked because one of the yielding vehicles departed after 10 s, all departures were detected except one. That one failure was in the treatment condition.
In the treatment condition, participants also detected that both lanes were blocked more quickly than in the control condition (F (1, 6) = 26.4, p < 0.01). The mean time to report that both lanes were blocked was 3.5 s in the control condition and 2.3 s in the treatment condition. The standard deviation in both the control and treatment conditions was 1 s. The sound-strip condition accounted for 81 percent of the within-subjects variance.
Unexpectedly, participants did not appear to have more trouble detecting two vehicles stopping at the same time than they did detecting two vehicles stopping with an intervening delay. Table 2 shows the number of correct detections of both lanes being blocked by stopped vehicles when those vehicles arrived at approximately the same time. The control and treatment rates of simultaneous yield detection, 27.5 percent and 53.5 percent, respectively, are similar to the overall hit rates shown in table 1. As with the overall rates, the detection of simultaneous yields was significantly better in the treatment condition compared to the control condition (t (6) = 2.3, p < 0.05).
To this point, results have been provided for the detection of two vehicles stopped side-by-side at the crosswalk or stop bar. These data were emphasized because the focus was on the accessibility of double-lane roundabouts, and it was assumed that crossing when only one vehicle has stopped at a two-lane crossing increased the risk of not hearing the approach of a second vehicle. However, participants indicated when they detected a vehicle stopping and in which lane. A hit was scored whenever the participant correctly identified, within 10 s, the presence and lane position of a stopped vehicle.
A miss was scored if a stopped vehicle was not identified as stopped in the specified lane within 10 s. A false alarm was scored when a vehicle was identified as stopped in a particular lane when it was not. If participants indicated that a response was incorrect within 1 s of making that response, then the first response was ignored whether or not it was correct. When a stopped vehicle was associated with the wrong lane, a miss was scored for one lane and a false detection was scored for the other. This was the case with approximately 19 percent of the misses in the control condition and 13 percent of the misses in the treatment condition.
Some participants incorrectly reported stops in a particular lane more than once within a trial (i.e., multiple false alarms). Multiple false alarms were possible in the single-lane case because participants could recognize that a vehicle was not stopped in that lane and could then falsely identify a subsequent stop. This was not the case in scoring for detection of both lanes blocked because in that scoring, a trial ended as soon as the participant indicated that both lanes were blocked. Because of the complexity of false alarm interpretation in the single-vehicle case, only hit rates are reported, and these should be interpreted with caution.
Figure 6 shows the proportion of correct identifications of stopped vehicles by condition, lane, and the lane in which the first yield occurred. Overall, performance was better in the treatment condition than in the control (F (1, 6) = 15.4, p < 0.01).
Identification tended to be more accurate when the first yield was in the far lane (F (1, 6) = 12.3, p < 0.05). The first vehicle to yield creates noise as it idles at the crosswalk. When it idles in the near lane, that noise can mask the later arrival of a second vehicle in the far lane. Participants were able to detect only 30 percent of the vehicles that stopped in the far lane when there was a vehicle already idling in the near lane. When the first vehicle to yield is in the far lane, its idling is somewhat attenuated by its greater distance from the participant, and the participant can usually detect the arrival of a second vehicle that comes between the participant and the vehicle in the far lane. When a vehicle yielded in the near lane, detection of which lane the vehicle was in was better if there was already a vehicle stopped in the far lane.
Overall, performance in the detection of stops was better for vehicles in the lane nearest the participant (F (1, 6) = 8.2, p < 0.05); however, these effects were not additive, since the three-way interaction of condition, lane to yield first, and lane was significant (F (1, 6) = 10.4, p < 0.05). In the control condition, detection performance for near-lane yields was affected very little by the presence of a stopped vehicle in the other lane. Conversely, in the control condition, detection performance for far lane yields was very sensitive to whether there was a vehicle already stopped in the near lane. The worst performance was for detection of vehicles stopping in the far lane when there was a vehicle idling in the near lane.
After completing the test procedure, participants were debriefed. The debriefing sought impressions of the differences in the two crossing conditions as well as additional information on the participants' experiences in making street crossings. The information obtained in these debriefings may be of interest to scientists and engineers who are concerned with accessibility.
Participant 1 is a middle-aged male. He takes walking trips outside his home or office every day. He travels alone during most of these trips. Many of these trips, almost every day, are to places he has not visited previously. He has some residual peripheral vision, but he relies primarily on hearing for street crossings.
Despite his superior performance in the sound-strip trials, he reported that the "seams in the pavement" made it more difficult to hear the car engines for which he has been trained to listen. He said that he consciously tried to ignore the noise from the seams. When he was told that his only false alarm occurred when a second vehicle drove over a strip when pulling up behind a vehicle that had already yielded, he appeared to be surprised. He said that he thought that he had entirely ignored the seams, but that perhaps he had not done so.
One issue in the accessibility of roundabouts is the amount of delay they impose on visually impaired pedestrians before these pedestrians identify an acceptable gap. In considering signalization versus other treatments, it may be worthwhile to consider the delay imposed by traffic signals. Participant 1 estimates that at signalized intersections, even those with which he is familiar, he is delayed by at least 4 minutes. This is because he always waits through at least one signal cycle to be sure he knows what is happening.
At some intersections where pedestrian call buttons are installed, he does not use them. His reason is that by the time he travels from the button to the crosswalk threshold and aligns himself with the crosswalk, the pedestrian phase has expired. At other crossings, he is unable to tell which crossing the accessible signal is indicating. In particular, he provided an example of an intersection he frequents where crossings are allowed only on one side of the intersection, but the accessible signal can be heard from both sides.
Participant 2 is an older (more than 65 years of age) male. He travels alone outside of his home or office every day. He estimates that about once a month he takes a pedestrian trip to a location that he has not been to before.
He indicated that he noticed the sound strips, but did not realize that they were intended to help detect yielding vehicles. He said that he tried to use the information these strips provided. He indicated that he did not know the number or location of the strips, but that such knowledge would have helped.
This participant said he sometimes crosses at uncontrolled crosswalks that have more than two lanes, even when only one lane is blocked. For instance, if it is otherwise quiet, he will walk in front of a vehicle that has yielded in the near lane so that he can better hear traffic in the far lane.
Participant 3 is a young adult female. She takes 6 to 12 pedestrian trips outside of her home or office every week. About half of these trips are with a sighted companion. She estimated that about once a month she takes a pedestrian trip to an unfamiliar location.
She indicated that she learned a lot during the test. For instance, she says that she learned that it is easier to detect vehicles in the near lane. She volunteered that the "bumps" made it harder to hear the car engines. She said that at uncontrolled crossings, if a car yields for her, she normally waves it on. Initially she said that she would never cross in front of a yielding vehicle if there were more than one lane. Later, she amended this by saying that if she could tell that there were no other cars, she would cross in front of a single yielding vehicle, but added that her experience with the test indicates that she could probably never tell if there were no other vehicles.
Participant 4 was a middle-aged female who said that she takes about 15 pedestrian trips outside of her home or office during an average week. Most of those trips are without a companion. She said that she travels to unfamiliar locations almost every week. She said she does not use visual cues when traveling, but that she can perceive some light, which is mainly useful at night.
She indicated that the test was difficult because she could not tell from which direction the cars were coming. She noticed the sound of the strips and said they were useful, especially when two vehicles arrived at the same time. When asked what she does when vehicles yield for her at uncontrolled crossings, she said she just waits until they leave and it becomes quiet. She does not signal drivers to move on, because she is unsure how they will react and cannot see them to adjust to their reactions. She avoids crossing at intersections without signals. She hates stopcontrolled intersections because she never knows what cars are going to do. That is, she does not know whether they have stopped for her or whether they will remain stopped if she decides to cross. She avoids streets with more than two lanes if they do not have signals. At signalized intersections, she always waits through at least one cycle so that she is certain about what is occurring. She prefers crossing midblock to crossing at nonsignalized intersections. When roundabouts were described to her, she remarked that the crosswalks were too close to the intersection; it would be better if the crosswalk were midblock and away from the noise of the intersection.
Participant 5 is a middle-aged male. He estimated that he takes about 10 pedestrian trips per week outside of his home or office, all without a sighted companion. Of these trips, perhaps two per month would be to locations that he had not visited previously. He indicated that he uses his residual vision to locate the white lines that demarcate crosswalks, and also to follow sidewalks.
Participant 6 is a middle-aged female. She reported taking about three pedestrian trips per week without companions, and walking to a location she has not visited previously about once per month.
She referred to the noise that the sound strips made as a "bubble wrap sound," because reminded her of the sound generated when her children pop the chambers in bubble wrap packing material. She stated that she preferred having the strips to not having them because were helpful in detecting when cars had stopped. She also indicated that she sometimes cross the two-lane street in front of her place of employment if vehicles stop for her.
Participant 7 is a middle-aged female who estimated that she makes one or two walking trips per week not near her house or office, usually with a sighted companion. She estimated that she takes walking trips alone to unfamiliar locations about six times per year.
She indicated that detecting the vehicles was easier with the sound strips than without them. She said that she would cross at crosswalks when she could hear car engines idling in both lanes. Although she was the only participant in this study who acknowledged having previously crossed at an actual roundabout, she said that she would not do that again on her own.
Study 1 indicated that a sound-strip installation could help some visually impaired pedestrians to detect yielding vehicles. The sound strips not only increased the probability of detecting stopped vehicles, but also decreased by more than a second the amount of time needed to make the detection.
However, in planning this study, a very low false alarm rate was set as a goal (i.e., a low rate of indications that a vehicle was blocking a lane when it was not). This goal was set to avoid a situation where the existence of sound cues would create a false sense of security. The false alarm rate in the control (without strips) condition was 10 percent. With the strips, the false alarm rate was 13 percent (essentially unchanged). This level of false alarms would not be acceptable for a deployable system.
Participants in this study were not trained to use the sound cues provided by the pavement treatment, nor were they informed of the treatment before the debriefing. It is conceivable that with training, detection performance with the sound strips would have been better, and false alarms might have been reduced. It is also possible that training would have no effect on performance, so the hypothesis that training would improve performance would need to be empirically verified.
Training was not included in the study because adopting a treatment that requires training would require an ambitious and expensive outreach. The most desirable treatment would be one that is self-explanatory, and this study's goal was to test whether the present treatment would work without training.
The finding that the detection of individual stopped vehicles was greatly improved with the presence of the sound strips suggests that the strips are effective in alerting participants to the presence of individual stopped vehicles. If only one lane of traffic had to be monitored, performance with the strips might approach that for right-lane detection in this study (80 to 90 percent correct detection). A sound-strip pavement treatment may be effective in single-lane roundabouts or right-turn slip lanes, even though the present results are not encouraging for the double-lane condition. The treatment has not been shown to work in the single-lane condition. The present results suggest that single-lane tests may be fruitful if some of the challenges identified in study 2 can be overcome.