• Federal Highway Administration >
• Publications >
• Research Publications >
• 09061 >
• 04.Cfm >
• Simulator Evaluation of Low-Cost Safety Improvements on Rural Two-Lane Undivided Roads: Nighttime Delineation for Curves and Traffic Calming for Small Towns

Simulator Evaluation of Low-Cost Safety Improvements on Rural Two-Lane Undivided Roads: Nighttime Delineation for Curves and Traffic Calming for Small Towns

PDF files can be viewed with the Acrobat® Reader®

Chapter 4. Discussion and Summary

Limitations of the Research

The current experiment was conducted in the FHWA HDS. This simulator was partially validated in the past for research on nighttime driving on rural two-lane curves by means of similar field data; however, these validations have been for different curve geometries and for different driver responses.⁽¹⁹⁾ In any case, there are inherent limitations to using simulator data to predict real–world driving responses. The simulator is often capable of producing similar relative results as those in the field for the purpose of ranking or comparing different treatments. This is one of the major strengths of using a driving simulator for highway safety research. However, when attempting to predict absolute quantities such as vehicle speeds or feature recognition distances, simulator–derived measurements may require scale factors or transformations before they can be used to estimate performance in the field.⁽²⁰⁾ In general, important relationships or patterns among data elements are usually preserved both in the simulator and in the field, but the simulator data often portray weaker and more variable results.

In the case of nighttime visibility research, there are two major reasons for such discrepancies. First, driving in a simulator is different from driving in a real car. Driver judgments of speed, distance, and deceleration are sometimes difficult in a simulator. Consequently, under certain conditions, drivers may drive faster and decelerate more rapidly in a simulator than in a real car. Also, there is no other traffic on the road. Second, the simulator has difficulty creating certain visual contrast ratios that can be encountered on the roadway, especially at night. The luminance for the visual stimuli used in this experiment (see table 3 ) are probably adequate for simulating pavement markings and possibly for simulating PMDs; however, the visual projection system of the simulator is not adequate to characterize the high contrast ratios found with streaming LED lights. Although these contrast ratios are high, the size of the LED light source is small, and the duty cycle is low, so glare should not be a problem for the driver. The inability to reproduce these high contrast ratios in the simulator was partially compensated for by increasing the size of the streaming LED light sources when they were far away. Nevertheless, there are distinct limitations to reproducing nighttime driving scenes in a driving simulator.⁽²⁰⁾

In the case of predicting the effects of daytime driving, the FHWA HDS has only been validated by means of similar field data for intersection traffic signals and for certain collision warning systems.⁽²¹⁾ The simulator has not been validated for traffic calming in small rural towns. However, from the stimulus perspective, the towns were presented only in daylight and had no intense luminous sources of light in the scene (e.g., traffic signals, tail lights, etc). The objects in the scene all represented reflected light. This is an easier type of driving scene to simulate and does not usually require high contrast ratios.

In view of the above limitations, absolute quantitative results from this experiment need to be considered with caution for both of the curves and the towns. However, the driving simulator is an excellent tool for exploring the relative differences among roadway treatments. This exploration of relative differences formed the focus of this experiment, and the observed relative differences may be regarded with a certain degree of confidence. As was pointed out previously, this confidence needs to be tempered with careful prudence to avoid other risks which are difficult to reproduce in a driving simulator, like the safety hazards posed by drivers and passengers exiting from the left-side doors of parked cars. For these and other reasons, field validations are usually warranted before simulator findings are recommended for implementation.

Curves

Summary of Findings for Curves

Four responses were measured for curves: speed, acceleration, curve direction detection distance, and curve severity detection distance. Table 9 summarizes the findings for three of these measures. The acceleration data were not summarized because they were derived from the speed data and portrayed similar relationships. Based on speed measurements at the point of curvature, columns 2 and 5 show the estimated average speed reductions for the various curve visibility improvements (in mi/h relative to the baseline rounded to the nearest 1 mi/h (1.61 km/h)) as well as the corresponding speed reduction rankings (from best to worst) for each.

Table 9. Estimated safety advantages and rank ordering of treatments for curves.

1 ft = 0.305 m1 mi = 1.61 km

Drivers performed almost perfectly in detecting the direction of curves ahead in the road (99.7 percent correct). In table 9 , columns 3 and 6 show the estimated average curve direction detection distance increases for each treatment (relative to the baseline rounded to the nearest 5 ft (1.52 m)) as well as the corresponding detection distance rankings for each. The 25-ft (7.6–m) average detection distance increase (11 percent) for adding edge lines in this experiment was almost identical to the 24–ft (7.3–m) average increase (12 percent) found in an earlier simulator experiment conducted by Molino et al.⁽⁸⁾ This correspondence was found despite the fact that the two simulator experiments employed curves of different radii and deflection angles. However, the absolute average detection distances were about 39 ft (11.9 m) greater in the current experiment. Such an outcome might be expected. Although the same simulator was employed, the graphics had been upgraded since the earlier experiment.

The absolute average direction detection distance for the single side PMDs condition in this experiment was 426 ft (130 m) (see table 7 ). Although under somewhat different conditions, a field study conducted by Turner et al. obtained an average curve detection distance of 656 ft (200 m).⁽²²⁾ A larger detection distance in the field might be expected for two reasons. First, the simulator was not able to produce the full visual contrast ratio experienced at night in the real world. Second, the field study used a 2–s search time from a stationary vehicle, which moved progressively closer to the curve in 100–ft (30.5–m) increments instead of a dynamic driving scenario which was used in this experiment.

Compared to curve direction detection, drivers did not perform as well on curve severity detection (only 82.5 percent correct). While this performance was better than chance, drivers had more difficulty detecting the severity of a curve than its direction. In table 9 , columns 4 and 7 show the estimated average curve severity detection distance increases for each treatment as well as the corresponding detection distance rankings for each.

A comparison of findings across the three response categories in table 9 reveals two important relationships. First, if the standard reflectorized PMD conditions were collapsed, there was a consistent order across all three response categories. Enhanced PMDs with streaming lights performed the best, the standard reflectorized PMDs performed the next best, and the edge lines alone resulted in the weakest or no improvement. Second, the enhanced PMDs with streaming lights performed exceptionally well in terms of increasing curve feature detection but only moderately well in terms of slowing drivers down.

Answers to Research Questions for Curves

Based on this experiment, the research questions posed in the introduction are answered as follows:

• How do the different visibility treatments perform to slow drivers down? In order of effectiveness, the treatments were organized into two groups: (1) delineators with edge lines and (2) pavement markings (edge lines alone). The delineators achieved a minimum speed earlier at the PC and achieved a greater overall speed reduction. By contrast, the pavement markings achieved a minimum speed later (at the middle of the curve) and achieved a lower overall speed reduction.
• What is the order of the four different visibility treatment conditions in terms of their effectiveness in slowing drivers down? Table 9 shows the observed order. If driving speed at the PC is used as the primary measure, the effectiveness rankings have the following order from best to worst with the estimated speed reduction shown in parentheses: streaming PMDs (9 mi/h (14.5 km/h)), both sides PMDs (8 mi/h (12.9 km/h)), single side PMDs (7 mi/h (11.3 km/h)), and edge lines alone (2 mi/h (3.2 km/h)). All of the PMD conditions contained edge lines as well as PMDs.
• Is there an effect of right versus left curve on driver speed profiles? Across all treatments, on average, the driving speed through the left curves was 1 mi/h (1.6 km/h) faster than through the right curves.
• Is there an effect of sharp versus gentle curves on driver speed profiles? Across all treatments, on average, the driving speed through the gentle curves was 10 mi/h (16 km/h) faster than through the sharp curves.
• Is there an adaptation effect? There was no adaptation effect.
• Do drivers slow down or speed up across the three experimental driving sessions? On average, the drivers did not slow down or speed up across the three experimental sessions.
• How do the different visibility treatments affect driver acceleration? The acceleration profiles for the curves organized themselves into two groups of treatments: pavement markings and delineators. Both groups of acceleration profiles had a sharp V–shape overall. However, the delineators were associated with a smoother and earlier deceleration pattern than the pavement markings.

The research questions and answers concerning feature detection for curves are as follows:

• How well do drivers perform on curve feature detection for the different visibility treatments? The drivers performed considerably better than chance on feature detection across all visibility treatments. The overall percentage of correct curve direction detection was 99.7 percent. The overall percentage of correct curve severity detection was 82.5 percent.
• How do the different treatments affect the distance from which curve direction and severity are detected by drivers? The different visibility treatments had a strong differential effect on improving curve direction detection distance. The treatments had less of an effect on improving curve severity detection distance. Average curve direction detection distance was organized into three groups: (1) streaming PMDs with edge lines, (2) conventional PMDs with edge lines, and (3) pavement markings (edge lines alone). Average curve severity detection distance was organized into two groups: (1) streaming PMDs with edge lines and (2) all other treatments.
• What is the order of the four different visibility treatment conditions in terms of their effectiveness in improving curve feature detection distance? Table 9 shows the observed orders. From best to worst, the following curve direction detection distance rankings were found with the estimated detection distance increases relative to baseline given in parentheses: enhanced PMDs with streaming lights (1,065 ft (325 m)), standard PMDs only on the far side of the curve (200 ft (61 m)), standard PMDs on both sides of the curve (130 ft (39.6 m)), and standard 4–inch (101.6–m) edge lines alone (25 ft (7.6 m)). As concerns curve severity detection distance, the corresponding rankings were found as follows: enhanced PMDs with streaming lights (560 ft (170 m)) and standard reflectorized PMDs (25 to 45 ft (7.6 to 13.7 m)). The standard 4–inch (101.6–m) edge lines alone performed the worst, no better than the baseline. It must be recalled that all of the PMD conditions in this experiment contained edge lines as well.
• Does feature detection distance change across the three experimental driving sessions? Is there an effect of multiple exposures? Feature detection distance consistently increased across the three experimental driving sessions for both curve direction detection and curve severity detection.
• Is learning a factor in interpreting novel visibility treatments? There was a strong adaptation or learning effect during the course of the experiment with regard to curve feature detection for the streaming PMDs condition.
• What is the effect of providing drivers with information regarding the curve direction and severity cues for the streaming PMDs condition? The drivers tended to learn the intuitive curve direction cue on their own. Direction detection distance did not increase substantially over projected learning after the drivers were told the meaning of the curve direction cue. The drivers had difficulty learning the less intuitive curve severity cue on their own. Severity detection distance increased substantially over projected learning after the drivers were told the meaning of the curve severity cue.
• Is there an effect of right versus left curve on feature detection distance? For both curve direction and severity detection distances, the effects of the direction of the curve were not statistically significant.
• Is there an effect of sharp versus gentle curves on feature detection distance? For curve direction detection distance, the effect of the severity of the curve was not statistically significant. However, the sharp curves had a greater average curve severity detection distance than the gentle curves (222 ft (67.7 m) versus 158 ft (48.2 m)).
• What is the relationship between slowing the driver down and improving feature detection on curves across the four treatment conditions? Drivers did not begin to substantially reduce their speed until they were about 200 ft (61 m) ahead of the curve. Yet, the enhanced PMDs with streaming lights resulted in curve feature detection distances of over 1,000 ft (305 m) ahead of the curve. As a result, drivers were aware of important curve characteristics (perception) far ahead of the curve but did not slow down (behavior) until much closer to the curve. An inverse relationship appeared in the case of the standard reflectorized PMDs conditions (without streaming lights). The both sides PMDs condition had a lower average direction detection distance (383 ft (117 m)) than the single side PMDs condition (466 ft (142 m)). This outcome was the opposite of what was observed for vehicle speed reduction where the both sides PMDs condition slowed the drivers down more than the single side PMDs condition by about 1 mi/h (1.6 km/h). The additional visual stimuli in the both sides PMDs condition may have confused the driver's judgments of curve direction and severity, but the formidable array of cues on both sides of the road, although confusing, may have served as a stimulus to slow down.

Potential Safety Benefits for Curves

As indicated above, the results of this experiment are more likely to be valid for rankings and relative comparisons among different roadway treatments than for absolute determinations of speed reductions or increases in detection distance. In the case of navigating curves on rural two–lane roads at night, safety improvements were defined in terms of reducing driving speed before and in the curve and increasing curve feature detection distance. The results of the experiment indicated that edge lines offered a small potential safety benefit, and standard reflectorized PMDs with edge lines offered a somewhat greater benefit. In general, standard reflectorized PMDs with edge lines performed better than pavement markings alone. This result does not imply that edge lines are not needed. Edge lines are still needed to provide continuous delineation of the travel lane, especially at close range. As concerns whether reflectorized PMDs should be implemented on the far side of the curve only or on both sides of the curve, the results were mixed. Since standard PMDs are considered to be a low–cost safety improvement, implementing PMDs on both sides of the curve is not likely to represent a large incremental expense. Thus, implementation on both sides of the curve should be preferred. Of all the treatments explored, the streaming PMDs with edge lines offered the most potential safety benefit by far.

Novel Curve Treatment Solution

For both curve direction and severity, adding streaming PMDs to standard edge lines resulted in a dramatic increase in detection distance. Without being told the meanings of the visual cues coded in the novel streaming lights, drivers were aware of the direction of the curve ahead and, to a lesser extent, of the severity of the curve ahead well before the curve itself came into view.

Since the streaming lights moved in the direction of the curve ahead, the direction cue was intuitive. This curve direction cue was learned quickly and well. The severity cue was less intuitive. Drivers had more difficulty learning the meaning of this cue. However, once they were informed of the meaning of the severity cue by means of verbal instructions, drivers could detect curve severity almost as far away as curve direction, although their accuracy was not as good. Training and/or education may be required before implementing such a curve severity coding scheme.

Although drivers were aware of the existence of a curve and some of its features at a great distance before the curve, they did not start to substantially slow down until they were much closer to the curve. When they did slow down, drivers slowed the most for the streaming PMDs, but the relative speed reduction was not as dramatic as the increase in feature detection distance. This outcome does not necessarily imply that the drivers forgot the perceptual information or were unprepared to act on it. Despite advanced information, it would be expected that drivers would still try to reduce total trip time by maintaining their speed until they came closer to the curve. Moreover, with advanced information on curve direction and severity, drivers would presumably be better prepared to act appropriately in an unexpected emergency situation.

Although the streaming PMDs solution is not yet technically mature for two–lane rural roads, future research might reveal practical implementation options. If optimal implementation strategies were selected, and the resulting system gained widespread adoption, the preinstallation cost could possibly be kept low. While not likely to become an extremely low–cost treatment, for instance when compared to standard PMDs, such a streaming light PMD countermeasure could become comparatively inexpensive with time, especially relative to modifying curve geometry at locations with high crash frequencies. Unfortunately, an exploration of the potential future cost implications of implementing the streaming PMD countermeasure was beyond the scope of this experiment.

Recommendations

The results from the current experiment identified one potentially low–cost safety solution worthy of further study and consideration—reflectorized PMDs enhanced by streaming LED lights. Several recommendations are suggested on how to proceed with this streaming PMDs solution.

Optimal Light Patterns

Further experimentation needs to be conducted to investigate optimal sequential flashing patterns for streaming PMDs. As a part of the preliminary pilot study for the current experiment, some initial exploration was conducted on this issue. Three different overall patterns were tried: (1) lights streaming toward the driver, (2) lights simultaneously flashing, and (3) lights streaming away from the driver. The latter pattern proved the most effective, producing an intuitive cue for curve direction. However, the cue of streaming cycle rate to indicate curve severity was chosen with little experimentation. A different flashing pattern, still in the direction of the curve, may have been more effective as a cue for curve severity. Even if the streaming cycle rate proved effective, the particular rates used in the current experiment were found to be discriminable but not necessarily optimal. Furthermore, in this experiment, there were only two curve radii (severities) and one deflection angle. A suggested future simulation experiment might employ more variation in curve radii and a wider range of curve deflection angles to ensure sufficient generalization across the full spectrum of field implementations and corresponding coding schemes.

Advanced Information

Further experimentation needs to be conducted to investigate the behavioral effectiveness of providing advanced information about the characteristics of curves far ahead in the road. In the current experiment, at a great distance before the curve, drivers were aware of the existence of a curve and some of its features. However, they did not start to substantially slow down until they were much closer to the curve. In addition to producing a modest decrease in driving speed upon curve entry, this advanced information may have an important secondary benefit. With advanced information on curve direction and severity, drivers may be better prepared to act appropriately in an unexpected emergency situation. This hypothesis might be tested in a future driving simulator experiment. The driving simulator is well suited to testing drivers' reactions to emergency events with possible serious negative consequences.

Technology Development

The feasibility of implementing a streaming PMDs solution for curves on rural two–lane roads needs to be explored, and the cost of possible implementation is an important factor for consideration. The streaming light technology for rural two–lane curves is not yet mature, although it has some application precedent for limited access roadways in other countries.⁽⁸⁾ In order to make this technology practical for remote rural locations and to reduce overall life–cycle costs, this technology would probably need to be implemented by solar power. If an array of streaming PMDs were connected together by underground wires, the entire array could be powered by a single solar panel and be sequenced by signal wires from a single control box. If each PMD were independent and self–sufficient, each one could be powered by its own small solar collector and sequenced by a system of radio frequency control. In this case, no underground wires would be needed, but the entire array would require exposure to adequate sunlight. In either case, in order to keep power consumption low, the streaming array could be activated by approaching traffic by sensing vehicle motion, headlights, noise, or other characteristics. The array could be made fault tolerant by automatically adjusting the streaming pattern of lights if a particular PMD in the array failed.

Field Studies

If the streaming PMDs solution seems feasible and cost effective, an experimental system needs to be constructed and tested both on a closed test track and later on a public roadway. The simulator data from this experiment need to be compared with field data collected under similar circumstances. The reductions in curve driving speed and increases in curve feature detection distances obtained in the laboratory need to be confirmed in the field. A closed test track experiment might be conducted using an experimental system of streaming PMDs and an instrumented vehicle. This suggested experiment might employ curves of different radii and deflection angles. Such an experiment should employ response measures similar to those used in this experiment: driving speed, longitudinal acceleration, curve direction detection distance, and curve severity detection distance. Data on lane position and lane excursions would also be useful.

If the results of such an experiment seem promising, a field test might be conducted on a public road outfitted with the experimental system of streaming PMDs. This experimental system should be applied on a few selected curves for a period of several weeks or months. A before/after/before field test might be devised. Such a field test might measure driving speed and deceleration on the selected curves before the experimental implementation of the streaming PMDs countermeasure has been put in place, during such an implementation, and after the implementation has been removed. The successful outcome of such a field test might reveal a baseline speed (and lane position) through the curve before the treatment, a significant decrease in speed (and possible improvement in lane keeping) during the treatment, and a recovery of the original baseline conditions after treatment removal.

Towns

Summary of Findings for Towns

The major findings regarding the relative advantages of the various speed–calming treatments are summarized in table 10. For both the beginning and middle of town locations, the table shows estimated average speed reductions (relative to the baseline condition) of the tested safety improvements as well as their rank ordering (from best to worst).

Table 10. Estimated speed reductions and rank ordering of treatments for towns.

1 mi = 1.61 km

Answers to Research Questions for Towns

Based on the findings of this experiment, the research questions posed in the introduction are answered as follows:

• How do the different traffic–calming treatments perform to slow drivers down? In order of effectiveness, the treatments were organized into three groups: (1) chicanes, (2) parked cars, and (3) baseline and bulb–outs. Except for the chicanes, the minimum speed was achieved near the middle of the town or slightly further into the town. For the chicanes, the minimum speed was achieved at the beginning and end of the town where the chicanes were located. In the middle portion of the town, the speed increased, creating a local peak in the shape of the speed profile functions for the chicanes.
• What is the order of the five different traffic–calming treatment conditions in terms of their effectiveness in slowing drivers down? Table 10 shows the orders observed. When driving speed at the beginning of the town was used as the primary measure, the effectiveness rankings had the following order from best to worst with the estimated speed reduction shown in parentheses: curb and gutter chicanes (9 mi/h (14.5 km/h)), painted chicanes (6 mi/h (9.7 km/h)), and parked cars (4 mi/h (6.4 km/h)). When driving speed at the middle of the town was used as the primary measure, the effectiveness rankings had the following order: curb and gutter chicanes (5 mi/h (8 km/h)) and parked cars and painted chicanes (4 to 4.5 mi/h (6.4 to 7.2 km/h)). Both types of bulb–outs produced either a small reduction in speed (about 1 mi/h (1.6 km/h)) or did not perform significantly better than the baseline.
• Is there an adaptation effect? There was no adaptation effect.
• Do drivers slow down or speed up across the three experimental driving sessions? On average, the drivers did not slow down or speed up across the three experimental sessions.
• How do the different traffic–calming treatments affect driver acceleration? The acceleration profiles for the towns were organized into two groups of treatments: (1) chicanes and (2) all other treatments with one exception for the parked cars condition. The parked cars condition revealed somewhat lower deceleration values before the town. The acceleration profiles for the chicanes had a pronounced jagged saw–tooth shape in the middle. This saw–tooth pattern represents rapid deceleration to navigate the first chicane at the beginning of the town, acceleration through the first part of the town, and somewhat less rapid deceleration to navigate the second chicane at the end of the town.
• Do some of the treatments slow drivers down before reaching the town while others slow drivers down only inside the town? All treatments slowed the drivers down before reaching the town. The chicanes tended to slow the drivers down the fastest and to the largest degree just before reaching the town. All of the other treatments tended to slow the drivers down the most while inside the town. In addition, the parked cars were associated with a more steady decrease in speed beginning further away from the town when compared with the other treatments.

Potential Safety Benefits for Towns

In the case of traffic calming for rural towns, safety improvements were defined in terms of reducing driving speed at the beginning and in the middle of the town. The results indicated that bulb–outs offered a small potential safety benefit or no benefit at all. Painted chicanes and parked cars on both sides of the road offered a greater benefit, and curb and gutter chicanes offered the most potential safety benefit. In general, chicanes and parked cars performed the best as traffic–calming countermeasures. Even the painted version of the chicanes performed well. This latter speed–calming measure is both effective and low in cost. If some parking spaces were eliminated at the entrances to the town, painted chicanes could be applied to slow drivers down. Flexible delineator posts could be added to the painted chicanes to reduce the tendency for drivers to cut the corners of the painted curves, possibly resulting in even greater speed reductions.

Alternatively, if the street were sufficiently wide, adding and encouraging parking on both sides could be implemented without any curb and gutter modifications. Thus, the parked cars solution could also prove to be both effective and low in cost. Although effective, curb and gutter chicanes would probably be more expensive than either painted chicanes or parked cars. The painted bulb–outs, while cheap, offer little or no advantage in terms of speed reduction. The curb and gutter bulb–outs would be expensive for an equally minor potential benefit.

Previous | Table of Contents | Next