bul 2004 
and the 
(3) 
linearly 
viewer's 
'd depth 
display 
display 
ng zone 
distance 
ng zone, 
side the 
z-plane. 
side the 
(4) 
ants, the 
wiable p 
©) 
d depth 
ssuming 
xe 5 for 
EE 
25 
rallax 
nt on the 
ot linear. 
parallax; 
viewing 
ed under 
s small 
perty of 
parallax 
hange of 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part BS. Istanbul 2004 
4. AUTO3D TOOLKIT 
The display device used in this study is the 2018XL 
autostereoscopic monitor manufactured by DTI Inc., USA. It is 
2D and 3D compatible and uses backlight technique to generate 
a sequence of light at certain frequency. Its maximum display 
resolution is 1280x1024 pixels. Unlike other popular 
autostereoscopic monitors, this monitor supports only two 
channels; therefore, the resolution is higher than other multi 
channel displays. A summary of the main specifications of the 
DTI monitor used in this study can be found in (Shan et al, 
2004). 
In order to evaluate the performance of autostereoscopic 
measurement, we developed a toolkit Auto3D based on the DTI 
autostereoscopic monitor. Auto3D is developed using 
Microsoft Visual C++ 60 with Multiple Document 
Interface/View frameworks. It can load, display and manipulate 
two images, conduct autostereoscopic measurement, label, and 
finally export the results. Figure 6 presents the main image 
measurement windows of Auto3D. 
     
    
Left Right 
monoscopic monoscopic : : 
Left image view view Right image 
À Fmt Hs 
  
The 3D viewing zones of this autostereoscopic monitor are 
created by parallax barriers. As discussed earlier, the principle 
of barrier-based system requires the two images of a stereo pair 
displayed being interleaved in columns. This indicates the 
horizontal resolution of the stereoscopic view is only half of the 
vertical resolution. Therefore, it is necessary to resample the 
two images properly to obtain both correct and sharp 
stereoscope. For this requirement, we duplicate the rows of the 
two original images for high quality application. Although this 
essentially doubles the image size for 3D display, as a trade-off 
the full resolution of the original images is retained. Moreover, 
the objective of image measurement is to obtain the image 
coordinates of feature points, such as corner point, line 
intersection, or T-junction, which should therefore be easily 
identified on the images. Consequently, Auto 3D is designed to 
handle two full-resolution images of a stereo pair. 
The 3D measurement in Auto3D is based on dual floating 
marks. Unlike many other digital photogrammetric systems, 
these two floating marks need to be an ellipse with major axis 
in the vertical direction. In this way, the interleave process in 
DTI monitor will create one circular cursor under 3D mode. 
This dual design also applies to any graphic interface that is 
desired to be viewed as 2D. This property may essentially 
double the work of software design and development (Shan et 
al, 2004). 
For data collection, Auto3D can currently digitize point 
features on the images. Properties of labeled points can be 
changed, colored, stored into a data file, and later loaded for 
either adding new measurements or editing previously existent 
measurements. Furthermore, with two cursors on the left and 
right views, Auto3D's internal frames can simultaneously 
display two images and their corresponding two pairs of 
monoscopic views in one document as shown in Figure 6. Each 
pair of monoscopic views includes individual left and right 
images. To start measuring the coordinates of features points, 
move these two images toward or away from each other by 
rolling mouse wheel to adjust x-parallax until the feature points 
observed under the 3D condition have the best stereo 
perception. Checking the small monoscopic views, we can 
confirm that both cursors on the left and right images are 
located on the identical position of the feature point and obtain 
accurate height information. 
5. TESTS AND EVALUATION 
5.1 Tests data and equipments 
Tests are designed to evaluate the performance of 
autostereoscopic measurement by comparing the results from 
different operators and from different equipments. To do so, a 
stereo pair at scale 1:4000 are scanned at a resolution of 33-um 
pixel size. Then, they are first epipolar normalized to remove 
possible y-parallax. The normalized images are resampled to 
two different resolutions, one at 25-um pixel size and the other 
at 50-um pixel size, which are used as the test images in our 
study. Two types of well-defined feature points are selected: 18 
points on the ground and 18 points on building roofs. Seven 
geomatics engineering major graduate students without 
intensive stereoscopic training are involved as operators in the 
study. The test organizer requests that all operators measure the 
36 feature points at two resolutions (25-um and 50-um pixel 
sizes) by using Auto3D toolkit and common photogrammetric 
workstation. During the measurement, the operators should 
follow the measuring specification prepared by the test 
organizer. In the specification, the exact location of each 
feature point is verbally described and illustrated with an image 
clip of 150x150 pixels. Figure 7 presents two of the feature 
points selected for measurement in the tests. 
     
   
  
   
  
  
Figure 7. Examples of selected feature points for measurement 
(left: ground; right: roof) 
  
  
  
For comparison purpose, a popular digital photogrammetric 
workstation is also used to conduct the same measurement. The 
software supporting the stereoscopic measurement on Windows 
system is Socet Set. The workstation equips with a regular 19” 
CRT monitor and requires operators wearing shutter glasses,