International Archives of the Photogrammerry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
which is NuVision 60GX stereoscopic wireless LC glasses
produced by MacNaughton, Inc. The alternating field rate of
this wireless LC glasses is 120 Hz (60 Hz per view). Under this
field rate, the left image and right images are alternately
displayed at one instant for each corresponding eye. During the
display interval, the LC-shutter blocks out each view by
providing time-multiplexing binocular parallax. For
synchronization, an infrared emitter is connected to the
computer that is operated for the left and right images from a
pair of stereo. This emitter generates a synchronization signal
that is decoded by the eyewear for accurately switching the LC-
shutters. In addition, a 3-pin connector into a compatible
graphic card and the eyewear is automatically activated
whenever a stereo application is running. The refresh display
rate of the graphic card can be accommodated to the most 120
Hz. While 120 Hz is the optimal rate, lower refresh rates are
entirely acceptable in order to accommodate a much wider
variety of display options. For our tests, we adjusted the display
resolution to 1024x768 and the refresh rate to 75 Hz.
To provide independent ground truth without potential side
effect of stereoscopic and autostereoscopic measurement,
reference measurement is prepared by the test organizer, who
uses Adobe Photoshop 5.0 to measure left and right images
coordinates of the selected features.
The study is primarily to evaluate the consistency of the
autostereoscopic measurements and comparing them with the
results from stereoscopic and monoscopic measurements.
5.2 Consistency of autostereoscopic measurements
This section will evaluate the consistency of autostereoscopic
measurements from different operators. Since seven operators
measure all the 36 feature points, the standard deviation can be
calculated respectively for x-left, x-right and y coordinates at
each feature points. Therefore, in total there are 3x36 standard
deviations calculated, respectively for the 25um and 50um
resolution images. Their distributions are plotted in Figure 8.
This figures indicate that the majority (> 85%) of the
inconsistency among operators is less than 2 pixels. About 15%
are within the range of 2-3 pixels. It should be noted that the
consistency of measurements tends to be dependent on the
image resolution. The lower the image resolution, the more
consistent (small standard deviation) the measurements in terms
of pixel size. This possibly suggests the resolution limitation of
the autostereoscopic monitor.
58%
(a) (b)
Figure 8. Distribution of standard deviations in pixels among
operators (a) 25 um. (b) 50 um.
5.3 Comparison with other measurements
The autostereoscopic measurements are compared with
stereoscopic and monoscopic ones. The type of test discussed
here is the paired / test. For the comparison of the differences
between two observations by using the autostereoscopic,
stereoscopic and monoscopic measurement, the paired ? test is
evaluated according
X aig
Pr I
Hy: yg =0 Hy: pty #0 Spin (6)
under the condition of rejecting Hy if i 214 /2.(n-1y » Where
Hair denotes the mean population difference, X diff is the
sample mean difference, S dify represents the sample standard
deviation of the difference, and n is number of points.
In the paired /-test, we compare every point measured for both
autostereoscopic (Auto3D) and stereoscopic (Socet Set)
systems to the point of reference measurement (Photoshop).
Moreover, every point measured with the autostereoscopic
system is also compared to the point from the stereoscopic
system. Such statistics are evaluated for the x-coordinate
measurements on both left and right image, respectively. Note
that since the images are epipolar normalized initially, the
values of y-coordinate for each feature point are specified as
the same. We calculate the differences between coordinate
values of the points measured by using different types of
systems for every identical feature point. A 95% confidence
interval is applied for mean difference here. If the confidence
interval for the combination contains zero and the p-value is
greater than 0.05, then the points are not statistically different.
Our statistical results are presented in Table 1 for the
resolutions of 25-um and 50-um pixel size, respectively.
A few analyses can be made on Table 1. The statistic exhibits
significant differences between autostereoscopic and
stereoscopic measurements for x-left observations under the
resolution of 25 uum and 50jum pixel size. The measurement for
50um x right observation has difference between
autosterecoscopic and stereoscopic systems. Notice that for y
observation the difference between autostereoscopic and
stereoscopic measurement shows significant difference under
the resolution of both 25 um and 50pm pixel sizes. The statistic
significant differences derived from the measurements for the
autostereoscopic system are generally larger than stereoscopic
and monoscopic measurements. Moreover, the results indicate
that under the resolution of 25um pixel size, the differences of
the autostereoscopic measurement are larger than that of
stereoscopic measurement for both x-left and y-observations. In
contrast, under the resolution of 50jum pixel size all results
obtained by using the autostereoscopic system are larger than
those from stereoscopic system. However, the maximum
differences between autostereoscopic and stereoscopic systems
in x coordinates for left image are 3.6 and 2.1 pixels under the
resolution of 25pm and 50um pixel sizes, respectively. The
corresponding maximum differences in y coordinates are 1.6
and 1.3 pixels. The maximum difference of either the x-
coordinate measurements for left and right image or the y-
coordinate measurements is less than 4 pixels under the
resolution of 25pm and 50pm pixel sizes.
Inte
Tal
<
=
do wb. |
a 13
x
61
th:
an
lo
an
