Keywords

1 Introduction

From 2014 to 2016, the number of people killed in motor vehicle collisions jumped from roughly 35,000 to over 40,000; in 2018, the number of traffic fatalities has remained at the 2016 level [1]. The increase in the death toll rate is somewhat surprising given the increased numbers of safety features found on new vehicles although these are available only in recent models. Even more surprising is the finding that, since 2010, cyclist fatalities increased by 25% and pedestrian deaths rose by 45% [2]. Possibly some of the increase in pedestrian deaths is due to increased emphasis on walking and biking in many cities. Nevertheless additional research and development is needed for improving the detectability of pedestrians, cyclists and other crossing traffic.

1.1 Heads up Displays

HUD technologies, due to their success in aviation, are being introduced in automobiles. It is important to examine how to design these displays so that information is displayed without interfering with information in the driver’s environment. Therefore, most investigations of driving performance with HUDs have focused on whether the projected image impairs driving performance. For example, Watanabe et al. [3] examined the effect of HUD on driving. Participants viewed a video of driving and were instructed to report when certain events (road signs, turn signals, and passing cars) were detected. Participants also detected triangles at varying locations that were projected on the HUD. Results showed that responses to warning triangles increased with eccentricity; however, the HUD task did not significantly interfere with detection of road events. Liu [4] examined whether HUDs affected driving performance by manipulating whether drivers attended to the HUD or the outside environment. Lui showed that drivers attending to the HUD responded faster to road events under both low and high driving loads. Moreover, variability in speed was lower in the HUD condition.

In one of the few tests of HUDs for pedestrian detection, Kim et al. [5] compared pedestrian detection performance in which pedestrian crossings were signaled either with a traditional “BRAKE” symbol or a virtual “Shadow” display that notified drivers of the pedestrian’s direction of travel. These were compared to pedestrian detection task with no cue. Both displays were shown to reduce braking reaction time and increase stopping distance. Therefore, HUD technologies may be useful tools in alerting drivers to objects in their periphery that are possible hazards.

The purpose of the present study was to examine the effects of visual warnings for obstacles that appeared in a driver’s periphery and crossed into the driver’s lane of traffic. We investigated whether the presence of a crossing alert affected driving performance, and three formats for signaling this event in a simulated HUD display. The Lane Change Task (LCT) and Peripheral Detection Task (PDT) were used [6].

2 Method

2.1 Participants

Twelve participants with at least 3 years driving experience were tested. All participants reported normal or corrected-to-normal vision and no color-vision deficits. Each participant completed two hours of testing and received a $20 gift card upon completion.

2.2 Apparatus and Materials

The LCT requires a participant to drive on an otherwise empty, three-lane road and change lanes in response to signs located on the shoulder of the road. The LCT was projected on a large screen, and participants completed the test using a Logitech gaming steering wheel. The LCT consists of 3-km straight-lane track and presents lane change signs roughly every 150 m, making a total of 18 lane changes per track. Performance on the LCT was assessed by the MDEV, the average deviation between the driver’s path and a normative model. The PDT stimulus consisted of a sequence of four small white circles each 200 ms in duration, giving the appearance of an 800-ms moving stimulus. The initial circle appeared in the participant’s peripheral view on the side of the road. The remaining circles appeared in sequence (ISI = 0 ms) and produced movement either toward or away from the center of the traffic lane. Participants responded to critical PDT events - movements toward the center lane, using a number pad located next to the steering wheel.

Four PDT conditions were tested based on whether critical events were cued, and the format of the cue, either color, flashing or color and flashing. For the color cue, the initial circle of the PDT event was yellow when movement was toward the center lane. For the flashing cue, the initial spot flashed when movement was toward the center lane. For the color-flashing cue, the initial spot appeared yellow and flashed for the critical event. For the flashing cue, the initial stimulus flashed twice. These cues were compared to a No-Cue Condition in which direction of movement was not cued.

2.3 Procedure

For each of the PDT cue conditions, three tracks each were run under low-workload or high-workload conditions in random order. For low workload, the PDT stimuli were presented between lane changes while the driver was lane keeping. For high workload, the PDT stimuli appeared while the driver was changing lanes. Within each track, nine PDT stimuli moved toward the center lane (critical events) and nine moved away from the road. Each participant was given time to practice the LCT prior to data collection. Baseline tracks with no PDT task, were run at the beginning and end of the cue conditions to determine the effect of the PDT on overall LCT performance. Driving performance was assessed with MDEV; performance on the PDT was assessed by reaction time and a non-parametric measure of sensitivity, A’.

3 Results

A one way repeated measures ANOVA was run to determine the effect of the conditions on driving performance (MDEV), with the factors baseline (no PDT), and the four cue conditions. The effect was nonsignificant (p = .582). A three-factor ANOVA was run on MDEV to determine the effects of cue conditions. Again, all main effects and interactions were nonsignificant (p’s > .39).

PDT performance was measured by A’, a non-parametric measure of sensitivity, and reaction time. A’ determines a participant’s sensitivity to signals in noise based on the number of hits and false alarms. A’ ranges from 0.5, indicating no sensitivity to 1.0 indicating perfect sensitivity. Overall sensitivity on the PDT task was high, as the mean A’ across conditions was .96. A three-factor repeated measures ANOVA with factors Color Cue (color vs. no color), Flash Cue (flash vs. no flash) and Workload (high vs. low) was run. Results indicated significant main effects of color, F(1,11) = 7.390, p = .020, np2 = .402, and workload, F(1,11) = 6.722, p = .025, np2 = .379. Sensitivity was higher for color cues, and for the high workload condition. The main effect of flash and all interactions were non-significant (p’s > .289). For reaction time, main effects of Color F(1,11) = 7.136, p = .022, np2 = .393, Flash, F(1,11) = 46.546, p < .001, np2 = .809, and workload, F(1,11) = 10.900, p = .007, np2= .498 were obtained. Color cues (M = 642.3 ms, SEM = 24.3 ms) reduced reaction time compared to no-color cues (M = 665.8 ms, SEM = 20.8 ms). Surprisingly, reaction time to Flashing Cues (M = 708.0 ms SEM = 25.3 ms) was higher than for non-flashing cues (M = 600.0, SEM = 20.6 ms), and reaction time for the low workload condition (M = 665.3 ms, SEM = 23.0 ms) was higher than for the high-workload condition (M = 642.7 ms, SEM = 21.9 ms). All interactions were non-significant (p’s > .096).

4 Discussion

These results indicate that cues to events in the driver’s periphery did not impair driving performance, at least for the simple driving task used here. For detection of critical events, color cues significantly reduced detection times, and color cues were more effective than either flashing or color-flashing cues. In fact, flashing cues disrupted detection somewhat as the mean reaction time was higher for flashing cues compared with non-flashing cues. This surprising result may have been the result of the brief presentation time of the PDT event. Given a total duration of 800 ms for the PDT event, the flashing stimulus appeared only for the first 200 ms, producing only 2 flashes. This brief flash event may have been insufficient to properly alert the driver.

Also somewhat surprising was the finding that detecting the peripheral stimulus was faster when the participant was engaged in making a lane change, yet performance on the lane change task was unaffected. It is possible that in the low workload condition the driver was anticipating the upcoming lane-change sign, which reduced sensitivity to peripheral events.