An exploratory study of driver eye scanning behavior on curves and straight roads
In this study, driver eye scanning behavior was studied on curves and straight roads on a winding two lane rural hilly highway under two illumination conditions - day and night, using an instrumented car with a corneal reflection technique television eye movement recording system. For the day condition, eye movements of two subjects were analyzed on seven curves and three straight level stretches of open highway from Athens to Shade and back. In the curve analysis, eye movements were analyzed separately for three sections (curve approach, curve section and after curve) for each of the seven curve environments, in each direction (North and South bound). For the night condition (with low beams), eye movements of one of the two subjects were analyzed for the same curves and level straight road stretches. Three of the seven curves and one of the three level straight stretches had life lites. The presentation of results include x-y eye fixation density and eye fixation duration density maps, eye fixation x-y averages and standard deviations, marginal x-y eye fixation and eye fixation duration distributions, eye travel distributions and out of view duration distributions. An object code analysis was performed for the curves and straight roads for the day and night conditions. Further, based upon four of the seven curves (curves A,B,C,D), estimates were obtained for the averages, standard deviations, ranges etc. of the following vehicle measures - lateral lane position of the vehicle, longitudinal, lateral and vertical acceleration, deflection of the gas pedal and speed. The eye movement analysis has indicated that roadway geometry influences some aspects of driver eye scanning behavior considerably. Besides, eye scanning behavior differences exist between the three curve sections. In the day condition for two subjects on curves A,B,C and D, one run in each direction on left curves (A,C - South bound: B,D - North bound) the mean fixation location (horizontal), which was almost straight ahead (AFPD=784 feet, µ=0.11°, ó=3.77°, n=105) in the curve approach zone, shifted by 2.27 degrees to the left in the curve section (AFPD=607 feet, µ=-2.16°, ó=3.82°, n=114) and was found to be statistically significant (t=4.42 for a=0.05). The mean fixation location (horizontal), shifted by 3.76 degrees from the curve section to the after curve (AFPD=387 feet, µ=1.60°, ó=3.45°, n=89) and was found to be statistically significant (t=7.34 for a=0.05). On right curves (B,D - South bound; A, C - North bound) the mean fixation location in the curve approach zone was to the right of the focus of expansion (AFPD=288 feet, µ=2.48°, ó=3.21°, n=102) and shifted by 2.11 degrees further to the right (µ=4.59°, ó=4.39°, n=115) in the curve section (significant, t=4.07 for a=0.05). The mean fixation location (horizontal), shifted by 5.53 degrees from the curve section to the after curve (AFPD=394 feet, µ=-0.94°, ó=4.15°, n=109) and was found to be statistically significant (t=9.69 for a=0.05). Though the eye movement patterns on left and right curves are not symmetrical, the shifts in the x-direction center of gravity, from the curve approach to the curve section, for both left and right curves is over 2 degrees. The shifts in the x-direction center of gravity from the curve section to the after curve for both left and right curves are about 4 to 6 degrees (due to entering a new curve). Curve driving is quite different from straight road driving and this is a good indication of the effects of roadway geometrics. The estimates of the mean fixation location (horizontal), for two subjects in the day condition on three level straight road stretches in both South (AFPD=1181 feet, µ=1.23°, ó=4.12°, n=111) and North (AFPD=724 feet, µ=0.35°, ó=3.82°, n=102) bound directions are in moderately close agreement (not significant, t=1.61 for a=0.05). The best estimate of the center of gravity for level straight road driving for both directions (AFPD=898 feet, center of gravity: x-dir. µ=0.81°, ó=3.90°, n=213; y- dir. µ=-0.25°, ó=1.68°, n=213) is quite different from both left (x-dir. µ=-2.16°, ó=3.82° y-dir. µ=0.37°, ó=1.73°, n=114) and right (x-dir. µ=4.59°, ó=4.39° y-dir. µ=0.55°, ó=2.13°, n=115) curves (curve section only). Besides, the visual spare capacity or 'free time' which the driver can afford is less on curves than on straight roads. The average number of fixations per 100 feet (3.56 fix. per 100 feet, averaged for left and right curves for curve sections only) is greater than that for a straight road (2.21 fix. per 100 feet) during the day. The x-direction marginal distributions for the number of fixations in the day condition for two subjects for four curves (A,B,C,D) in both directions for left and right curves were found to be statistically significant for all the three curve zones (In the K-S two sample test, D=maximum calculated difference, D o=critical difference; curve approach- D=0.37, D o=0.189; curve section- D=0.58, D o=0.17; after curve- D=0.315, D o=0.194 for a=0.05). The y-direction marginals were statistically significant only for the curve approach zone (D=0.215, D o=0.189 for a=0.05). Temporal eye movement results appear not to be affected by roadway geometry. For the day condition for two subjects on all curves (A,B,C,D,E,F,G) and on all level straight roads (#1,2,3) in both directions, the average fixation duration in seconds is slightly longer on straight roads (µ=0.387, ó=0.369, n=213) than on curves (µ=0.32, ó=0.28, n=460). However, these results were not found to be statistically significant (K-S two sample test D=0.08, D o=0.113 for a=0.05). The geometric features of the curve (start and end of radius) offer no indication about the instant at which the visual search patterns of drivers change while approaching and negotiating a curve. A detailed 100 feet by 100 feet analysis (2 subjects, day condition, for curves A,B,C,D, North bound, South bound and both directions) indicates that the driver gets into 'curve mode' 4 to 5 seconds (300-400 feet) before entering the curve (V=50mph approximately). The night condition results for one of the two subjects indicate that the average foveal preview distance is slightly less on left curves with life lites (AFPD=110feet, y-dir. center of gravity=-2.0°, ó=1.61°, V=39mph, n=18) than that on curves with no life lites (AFPD=140feet, y-dir. center of gravity=-1.57°, ó=2.25°, V=37mph, n=46) but was not found to be statistically significant (t=0.853, a=0.05). The average foveal preview distance is much greater on right curves with life lites (AFPD=274 feet, y-dir. center of gravity=-0.80°, ó=2.08°, V=39mph, n=25) than that on right curves with no life lites (AFPD=117 feet, y-dir. center of gravity=-1.88°, ó=1.53°, V=40mph, n=56) and was found to be statistically significant (t=2.32 for a=0.05). The shorter average foveal preview distance at night (AFPD=119feet, y-dir. center of gravity=-1.84°, ó=1.35°, n=113) when compared to the day (AFPD=385feet, y-dir. center of gravity=-0.57°, ó=1.64°, n=109) suggests that the driver has an increased need to fixate on visual cues on the road close in front of the car to maintain proper directional and lateral control of the vehicle, since during night the visual and especially the peripheral input is severely reduced. The fixation durations were found to be longer at night (µ=0.422, ó=0.349, n=204) on all curves for both directions than the durations during the day (µ=0.323, ó=0.203, n=208) and were found to be statistically significant (K-S two sample test, D=0.155, D o=0.134 for a=0.05). This might be attributed to the reduced amount of visual stimuli and more accomodation time needed at night when compared to the day condition. The longer fixation durations at night might also be an indication of the lower efficiency of the visual mechanism at night. The driver appears to fixate mostly on the road surface at night in the vicinity of the forward edge of the low beam pattern. For a single subject, under level straight road (#1,2,3, both directions) driving conditions, it is observed that the fixation duration in seconds is slightly longer at night (µ=0.458, ó=0.344, n=113) than that during the day (µ=0.392, ó=0.343, n=109) and is statistically significant (K-S two sample test, D=0.185, D o=0.182 for a=0.05). It should be noted that the average out of view duration in seconds at night (µ=0.458, ó=0.344, n=18) is considerably lower than that during the day (µ=0.664, ó=0.369, n=30), but the results were not statistically significant (K-S two sample test, D=0.265, D o=0.316 for a=0.05). The fixation durations are slightly shorter on curves with life lites (curves E,F,G, µ=0.348, ó=0.234, n=70) than on curves with no life lites (curves A,B,C,D, µ=0.459, ó=0.391, n=134). However, the results were not statistically significant (K-S two sample test, D=0.165, D o=0.20 for a=0.05). Similar results were obtained for the same subject under night condition on straight roads (#1,2- no life lites, #3- with life lites). The fixation durations are shorter on straight roads with life lites (µ=0.362, ó=0.206, n=46) than on straight roads with no life lites (µ=0.523, ó=0.402, n=67), but the results were also not found to be statistically significant (K-S two sample test, D=0.175, D o=0.260). The direct eye travel on straight road (#3) with life lites in all directions (µ=3.37°, ó=2.21°, n=23) was longer than on straight roads (#1,2) without life lites (µ=2.19°, ó=1.72°, n=47) and was found to be statistically significant (K-S two sample test, D=0.375, D o=0.346 for a=0.05). From the object code analysis results it can be stated that on curve sections in day time (curves- A,B,C,D, 2 subjects, direction all) the subjects spent an almost equal amount of time fixating on the road in front (36.38%) and the road environment (39.79%), while the night condition results (for one of the two subjects only) indicate that on the same curves the subject spent 60.92% of the time on the road ahead and 21.73% of the time on the surrounding road environment. During night, an appreciably greater amount of time is spent looking into the road environment in the curve section (21.73%) than in the curve approach (4.73%) or after curve (9.55%) due to the contiguous scenery in the curve (day time, two subjects: CA- 34.64%, CS- 33.88%, AC- 34.36% in the road environment). On straight roads (2 subjects, straight roads #1,2,3, all directions) during day, the subjects spent less time viewing the road surface (20.04%) than the road environment (37.39%), while during night, one subject spent an appreciably greater amount of time (54.48%) on the road ahead than in the surrounding environment (6.06%) probably due to the darkness of the road environment. The object code analysis for one subject on straight road #3 with life lites at night indicated that 56.52% of the total fixations were spent on life lites while in the curve study (curves E,F,G, direction all, curve section only) the same subject spent 27.91% of the total fixations on life lites (straight road: 0% left edge, 56.52% center, 13.04%, right edge; curves: 0% left edge, 26.9% center, 4.65% right edge). The averages, standard deviations and ranges of the various vehicle measures were estimated for each subject on four curves (A,B,C,D) during the day condition. From the analysis it can be tentatively stated that the inherent variability of individual drivers, is the single largest source of vehicle measures variability. The package of eye movement and vehicle measures data analysis programs that were developed for this study are self contained and easy to operate. An important design feature is that the three eye movement data analysis programs accept the same data files for the analysis, and in spite of the high storage requirements of two of the programs (EMA, ETA), the programs can be executed from an on line computer terminal. It should be stated that reducing eye movement and vehicle measures data from a T.V monitor is a time consuming process. On the average, it can be stated that reducing 10 seconds of driving eye movement data (averaged over day and night conditions, curves and straight roads) takes approximately 1.3 hours. Reducing 10 seconds of vehicle measures data on curves (sampling rate, every 10 frames) takes approximately 1.8 hours and is more time consuming than the eye movement analysis. The major conclusion from this day and night exploratory eye movement study is that the drivers appear to operate with a much shorter average foveal preview distance at night when compared to the day results (straight road: night AFPD=121 feet at 43.5 mph, day AFPD=952.5 feet or 7.87 times longer at 50mph; left and right curves, curve approach, curve section and after curve together: night AFPD=130.7 feet at 38.6 mph, day AFPD=576.7 feet or 4.41 times longer at 46.8 mph). The average preview time at night for straight road driving is 1.9 seconds while the average preview time for both left and right curves is 2.3 seconds. These average preview times are rather short for the safe operation of a car at night.
School Location:USA - Ohio
Source Type:Master's Thesis
Keywords:corneal reflection curve driving eye travel distributions
Date of Publication:01/01/1980