DATA SET 5: Mike Land's Car Driving Data Set

Description
Although there had been a number of studies of eye movements during driving prior to this study, they mostly involved freeways or other wide roads, and were not particularly challenging in terms of visuo-motor coordination. We (Dave Lee and I) therefore chose a narrow winding road - Queen’s Drive round Arthur’s seat in Edinburgh (Fig 1) – because it required steering to be under continuous visual control. We made simultaneous recordings of steering wheel angle and gaze direction using a head-mounted video-based eye monitor with a resolution of about 1º. Head movements relative to the car were also monitored. The most striking feature of the records were that on bends in either direction gaze sought out and tracked the ‘tangent point’: the moving point on the inner lane edge that protrudes into the road (Figs 2 & 3). Gaze could take ‘time out’ from monitoring the road (e.g.the jogger in Fig 2), but not for long, and not usually on bends. The significance of the tangent point seems to be that its location, relative to the current heading predicts the curvature of the upcoming bend, independent of distance, according to the formula:

curvature = 1/(bend radius) = = θ² / 2d,

where θ is the angle of the tangent point from the current heading (i.e. the gaze angle if the tangent point is fixated), and d is the distance of the driver’s head from the inner lane edge. Subsequent simulator studies (detailed in Land, 1998) showed that drivers monitor both θ and d simultaneously. Since at low-to-moderate speeds steering wheel angle is proportional to path curvature, θ and d provide the driver with a direct control signal for steering.

Fig 1: The drive shown in the videos begins at the entrance to Queen’s Drive (top left) and proceeds to the bottom of the picture (Note videos are left-right reversed). Picture from Google Maps.

Fig 2: Four frames from a video of a drive around Queen’s Drive showing typical gaze directions (white dot) on a left-hand and right-hand bend (top), a straight segment, and while looking off-road at a jogger. Note that the subject’s eye occupies the lower third of each frame and unlike the videos themselves the frames are not left-right reversed. The scrap of tape attached to the windscreen allows head movements to be monitored.

Fig 3: Contour plots showing the density of fixations (relative to the maximum value) made by three drivers for the whole route shown in Fig 1. The show all left-hand bends (top), straight segments, and right-hand bends. The central 1.0 maximum has the approximate value of 0.12 fixations deg^-2.s^-1. Note the strong peaks centred within 1º of the tangent points.

Method
Eye movement recordings were made with a head-mounted camera that produced a split image in which the top two-thirds showed the scene ahead and the lower third the eye in its socket, imaged via a concave mirror. The location and ellipticity of the iris were used to obtain the coordinates of eye direction, by matching the iris outline to a computer- generated eye model. This was done by hand, frame-by-frame, at 50 f.p.s. The coordinates were used to position a 1º dot on the upper scene view, and each frame re-recorded. Head movements could also be obtained by tracking distant background objects in the scene view. The resulting video contains numerical values (in degrees) of the direction of view of the fovea, a frame counter, and a clock. The videos are reversed left to right as a result of the mirror optical system.

References

SCENE/DISPLAY/STIMULI IMAGES/VIDEOS

SCANPATH (EYE-TRACKER) DATA

Driver A Preview (27668 KB, DIVX compression)

Driver A Full Data Set (uncompressed avi, zipped: 1872310 KB)

Driver J Preview (34118 KB, DIVX compression)

Driver J Full Data Set (uncompressed avi, zipped: 2351198 KB)

Driver M Preview (28662 KB, DIVX compression)

Driver M Full Data Set (uncompressed avi, zipped: 2237447 KB)