«A Doctoral Dissertation Proposal by BRADFORD BRIMLEY February 2014 Dissertation Proposal: Brad Brimley Page 1 of 20 INTRODUCTION The horizontal ...»
A THEORETICAL MODEL OF VISUAL ATTENTION TO PREDICT
DRIVER PERFORMANCE AT CURVES
A Doctoral Dissertation Proposal
Dissertation Proposal: Brad Brimley Page 1 of 20
The horizontal components of a roadway can be divided into two types of sections: tangents, and the changes in alignment, hereafter referred to as curves, which connect tangents. Both components are necessary: while tangents provide the most direct connection between two locations, curves allow the tangents to be arranged so the road avoids crossing critical areas. For the inattentive driver, a curve can be a potential hazard. Research has shown that in order to navigate a curve safely drivers must satisfy increased visual demands within the curve (1). But little has been done to show the importance of driver visual behavior on curve approaches, where drivers obtain important information about the upcoming curve. If the visual behavior of drivers is different on curves than on tangents, there should be some measurable change on an approach as drivers notice and then anticipate a curve.
The effects of a driver’s attention on the approach to a curve will likely be reflected in how well he navigates it. If a driver’s performance while navigating a curve is related to his preceding visual behavior or attention, it is important to encourage an increase in driver attention at an appropriate time before these critical locations. It is hypothesized that the visual attention exhibited by drivers in anticipation of a curve increases when traffic control devices (TCDs) are used. Since TCDs provide visual information that may not otherwise be obtained from the curve itself or its surrounding environment, it is thought that this stimulus alerts drivers of the upcoming hazard, leading to a change in visual attention. When attention increases, performance should improve.
This proposal describes a plan to evaluate how driver visual behavior changes near curves and identify the relationship between visual attention before curves and operational performance while navigating curves.
PROBLEM STATEMENTThere are three hypotheses that will be tested in the dissertation: (1) driver visual behavior on a tangent changes as the driver approaches a curve, (2) driver visual behavior can be influenced by the use of TCDs at the curve, and (3) driver visual behavior prior to a curve affects performance while navigating the curve.
Visual behavior, which is a reflection of attention, will be characterized by parameters measured by eye-tracking equipment. Navigational performance is characterized by vehicle operational measures that are derived from values of speed or acceleration. In this study, the speed and acceleration data will be measured continuously, which allow for the development of new metrics that may bemore descriptive of driver behavior than speed or acceleration at one location alone.
The TCDs that will be tested are post-mounted delineators and large arrows, each used separately but in addition to pavement markings and advance warning signs. If these three hypotheses can be successfully proven, the resulting implication is that the tested TCDs can lead to better driving performance because they support improved attention. This investigation can Dissertation Proposal: Brad Brimley Page 2 of 20 lead to improvements in how the effectiveness of a TCD is measured: if a particular device leads to increased attention, and an increase in attention leads to safer navigation, there is justification to use that device at more-severe hazards. At a curve, for example, the level of attention supported by a particular device should be consistent with the severity of the curve.
SPECIFIC RESEARCH OBJECTIVESThere are three objectives to the research that need to be met in order to test the three hypotheses. For the written dissertation, it is anticipated that each objective will be fulfilled in a
separate chapter. The three objectives are:
1. Using data collected from study participants, develop a model that characterizes how a driver’s visual attention changes as the driver approaches and navigates a curve and identifies how that attention is influenced by TCDs, if any, used at the curve. This objective tests the first two hypotheses.
2. Develop metrics of driving performance on curves that characterize driving behavior better than traditional operations-based measures. Evaluate the driving performance of the study drivers using these metrics. This objective alone does not test a dissertation hypothesis, but it is thought that the new metrics will be beneficial in testing the third hypothesis.
3. Using the data from Objectives 1 and 2, identify the components of a driver’s visual attention prior to a curve that affect performance while navigating the curve. Identifying this relationship will test the third hypothesis.
The scientific literature is filled with evaluations of driver performance under a number of circumstances and in many different settings. For curves specifically, the research question has usually been whether or not a particular geometric element or TCD results in a measurable improvement in terms of vehicle operations. Rather than viewing the problem solely as a matter of operations, however, there may be a more important issue related to driver attention and the relationship between attention and operations. The purpose of this dissertation is to establish that relationship, which would identify the real mechanism that makes TCDs effective by investigating processes that are more internal to the driver. This section of the proposal discusses aspects of driver behavior in terms of operations and visual attention, the use of TCDs on curves, the effects of TCDs on driver behavior, and connections between visual attention of drivers and their driving performance. The section concludes with a discussion of research needs that support the hypotheses of the dissertation.
Driver Behavior Though specifics in the literature vary by location and characteristic studied, curves experience higher crash rates than tangents (2–4). The prevalence of crashes at curves has long justified Dissertation Proposal: Brad Brimley Page 3 of 20 extra attention from researchers attempting to better understand driver behavior. The process of navigating a curve is more complicated than driving on a tangent because the driver is required to make steering and speed adjustments that fit with the changing road alignment. This section contains a discussion of operational and visual behavior relevant to the driving task as a driver approaches and navigates a curve.
Operational Metrics for Curve Driving
Speed has been identified as a principal factor in crash frequency and severity (5, 6), making it the primary measure of interest when evaluating safety on curves. From a simple design perspective, the “design speed” of a horizontal curve is controlled by the radius, side-friction, and superelevation of the roadway. Because unique drivers have different levels of experience and risk-taking habits, vehicles differ by design and type, and curves are unique based on local conditions that restrict their design, speeds on curves can be variable. Predictions of driver speeds on curves have been done many times using different geometric elements such as radius, lane width, and grade.
Lateral forces acting toward the center of the curve must be applied in order for a vehicle to travel in a circular path. Lateral acceleration has a quadratic relationship with longitudinal speed and an inverse relationship with radius (without considering superelevation). The lateral acceleration felt by the driver is sustained by sideways friction between the tires and pavement, typically referred to as the sidefriction demand. Design guidelines for horizontal curves are set conservatively so that the sidefriction demand should never exceed what is available. With many crashes at curves identified as run-off-road, however, the lateral acceleration experienced by drivers is still an important element of safe navigation.
Research shows that drivers do not adjust their speeds on curves to consistently accept the same amount of lateral acceleration. At higher speeds, drivers are more cautious and accept lower levels of acceleration (7–10). If approach speeds are high, which is especially common at isolated curves, and drivers generally do not accept high levels of lateral acceleration at high speeds, they will need to decelerate to a speed appropriate for navigating the curve. This process begins after the driver becomes aware of the curve and begins formulating a strategy to navigate it. Driver speed throughout a single curve is usually not constant. Generally, the minimum speed is reached near the midpoint of the curve. Speed prediction models by Poe and Mason (11), Donnell et al. (12), and Islam and Seneviratne (13) each estimate a lower speed at the midpoint than at the point of curvature (PC), and those that also modeled speeds at the point of tangency (PT) estimate an exit speed close to the entrance speed. Medina and Tarko (14) and Hu and Donnell (15) observed that 66 percent of the total deceleration occurs on the tangent preceding the curve, with the remaining 34 percent occurring after the PC.
Though not receiving as much attention as speed, the placement of vehicles within the lane is an important component of curve navigation. When possible, drivers tend to “cut” curves (also called “cornering” or “flattening”), adjusting their position within the lane in order to navigate the curve at a radius larger than that of the actual alignment. For left curves, the vehicle encroaches upon the centerline. For right curves, the vehicle encroaches upon the edgeline.
Dissertation Proposal: Brad Brimley Page 4 of 20 Cutting the curve is intentional because the driver sacrifices position within the lane to reduce forces of lateral acceleration without having to reduce speed.
The operational effects of curves as discussed in this section show that the decisions made by drivers near curves are both intentional and natural. This is important because the decisions drivers make regarding how they navigate a curve are based on the unique characteristics of their vehicles and driving habits, their previous experiences, and the information obtained about the curve, usually in advance of the curve. That information comes from both the visible road scene and surrounding environment, including the TCDs, if any, used. The information obtained by the drivers is dependent upon their behavior on the approach to the curve, defined by the attention the devoted to the forward scene.
Our understanding of the role of visual attention in the driving task has evolved as new technology provides researchers with better information about the interaction between drivers and the driving environment at a visual level. With external technology (e.g., cell phones) being used in vehicles, there has been a surge of discussion recently about distracted driving, its effect on safety, and how it can be reduced or eliminated. This dissertation does not deal with distracted driving per se (defining distracted as failing to exhibit attention), but rather identifying the situations where drivers may exhibit different levels of attention. It seems that the concept of attention and its effect on driver behavior are too complex to be reduced to a dichotomy of attention vs. inattention.
The most notable early research on driver visual behavior was performed over 40 years ago by researchers Mourant and Rockwell (16) who worked to identify the general location of driver gazes as they repeatedly drove the same section of road. The drivers were specifically instructed to view all of the traffic signs during the first phase of the experiment, and then fewer and fewer signs as they repeated the course. Naturally, Mourant and Rockwell were able to show that driver gazes are shifted up and to the right when they are unfamiliar with a road, as simulated by the participants intentionally looking at each sign. They produced the images in Figure 1 that show the location and concentration of the gazes based on percent time at intersecting coordinates.
Where the concentration is high enough, a whole percent is given. A single dot represents locations with less than a whole percent of gazes. Note that in the image for the third drive, there are values as high as 5, 6, and 7 percent at some locations.
Dissertation Proposal: Brad Brimley Page 5 of 20
Mourant and Rockwell later identified differences in visual patterns between novice and experienced drivers (17). The glances of novice drivers covered a narrower horizontal field than those of the experienced drivers. Also, the vertical components of the novice drivers’ glances were lower, indicating that their preview distances were shorter than those of the experienced drivers.
In 1977, Shinar and McDowell teamed up with Rockwell on a study of visual behavior of drivers navigating curves (18). They investigated how eye movements differ when drivers are on a straight road with no approaching visible curve, on an approach immediately before a curve, and on the curve. The researchers found that lateral eye movements generally follow the direction of the curve beginning 2-3 seconds before entering the curve. In terms of vertical movements, the eyes exhibited patterns of fixations far ahead of the vehicle followed by brief fixations near the vehicle, as if the driver needs supplemental verification of his position. Based on the fixations on the road and scenery while the driver is on approach tangents compared to fixations when on curves alone, the researchers concluded that the process of curve negotiation actually starts before the curve, indicating that the visual behavior on the approach is extremely important.
Also, drivers tended to fixate on the road more while navigating right curves than left curves, which supports the need to evaluate visual behavior on curves differently by direction.