The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
390 
Figure 9. Captured stereo images from 18 directions 
Model 1 Model 2 
Figure 10. Prepared 3-D data for Approach 2 
4.2.3 Approach 3: The gaze tracking procedure without 
a known 3-D model 
Four types of building diorama models are used as the material 
for this experiment. 
3-D data for the building diorama models are generated through 
the image subtraction algorithm, as shown in Figure 11. 
Û 
3-D data 
1 1 
Figure 11. Image subtraction procedure 
First, stereo images without the building diorama models are 
captured as background images. Next, stereo images including 
the building diorama models are captured as input images. 
Then, segments of the building diorama models are extracted 
through the image subtraction procedure, from the input images 
and background images. At this time, a median filter and a 
small segment filter remove minor noise, such as 
correspondence errors. Finally, the 3-D data for the extracted 
objects from one viewpoint are generated via stereo matching 
as shown in Figure 12. 
Figure 12. Prepared 3-D data for Approach 3 
4.3 Gaze tracking 
The experiment of gaze tracking is conducted as shown in 
Figure 13. An object rotates on the turntable, and the active 
stereo camera moves to track the object continuously. Sequence 
images are acquired in the gaze tracking procedure at 15 
frames/sec. Sequence images at 180-frame intervals are shown 
in Figure 14. 
Figure 13. Gaze tracking experiment 
Frame 60 Frame 1320 Frame 2580 
Figure 14. Sequence images in gaze tracking