a target with no vertical disparity change to drive the eyes will
invariably show a vertical eye position difference of no more than
2–3 arc min. This fact is useful in evaluating a binocular eye tracker,
because it provides a convenient benchmark for evaluating
tracker noise. If a binocular eye tracker shows a standard deviation
for the difference in vertical position of more than a few arc minutes,
it is almost certainly due to system noise (Bedell &
Stevenson, 2013). Common video eye trackers show noise levels
of around 15 arc min by this test, despite the fact that benchmarks
with artificial eyes may show much better precision.
4.3. Advantages and limitations of the binocular tracking SLO
Our primary objective in the development of this system was to
achieve binocular imaging and tracking from a single SLO system.
Although one might design a truly binocular system with independent
stimulating and detecting channels for each eye, the modification
we describe to a single SLO greatly reduces the cost and
complexity over a two system design. The modification we
describe does have some drawbacks: splitting the scan reduces
the field size for each eye; the added left eye channel has one extra
mirror in the path, which slightly reduces light levels; and alignment
of both eyes to one system is a challenge. These factors can
impact image quality, and thus increase noise. However, with high
quality optics and careful alignment, the system has the same performance
as the monocular system described by Sheehy et al.
(2012). Our noise level estimates here are in agreement with the
previous report.
Compared to other systems for precise eye tracking, such as
search coils, contact lens mounted mirrors, or dual Purkinje image
trackers, the tracking SLO has the advantage that the retina itself is
being imaged and tracked. The position of a target on the retina can
thus be visualized, lending high confidence to the accuracy of stabilization.
The system we describe does not yet stabilize targets on
both retinas in real time tracking, but this requires only a relatively
straightforward software modification to implement the split field
independent tracking.
The alignment of the system to both eyes is a significant challenge.
The standard practice with monocular imaging is to position
the subject’s head to align the eye with the table mounted optics.
The retinal location of interest is then controlled by having the subject
fixate points either inside or outside the imaging raster. We
follow this head positioning procedure for aligning one eye of
our subjects, but aligning the second eye requires adjustment of
the mirrors in the system to accommodate variable pupil separation
and head angle. In this first implementation we concentrated
our efforts on aligning the beam with the pupil to optimize the
image quality by adjusting the position and angle of mirrors
CM6l and FM4 (see Fig. 1). Aligning the rasters to also fall on corresponding
retinal loci proved difficult due to the interaction
between beam angle and pupil entry location as mirrors are adjusted.
For the measurements we describe here, the rasters always
appeared overlapped in the peripheral visual field but were not
precisely aligned to binocular correspondence. Future designs will
incorporate better controls on the mirror components to facilitate
orthogonal control of beam position (alignment in the pupil) and
beam direction (alignment on the retina) for both eyes, for real
time stabilization of targets with well-controlled binocular
disparity.
The tracking SLO design allows for stimuli to be presented in
the scanning raster, but also provides real time eye position output
signals. The implementation of real time tracking of both eyes will
allow stabilization of targets in secondary displays as one might do
with a search coil or Purkinje tracking system. In that case strict
alignment of the rasters to corresponding retinal locations is not
required.