The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B3b. Beijing 2008 
provides real-time processing for the object tracking and gaze 
control. 
In addition, when an object is outside the image field or beyond 
a threshold of camera view, a ‘camera displacement’ (see 
Figure 2) is executed to relocate the object of interest to the 
center of captured images by active stereo camera translation 
and rotation. The camera displacement outputs the repositioned 
object parameters for the images. Then, the object in the images 
is repositioned via model realignment. The repositioned object 
parameters are fed back from the camera displacement task to 
the identification task. 
4. EXPERIMENTS 
4.1 Overview 
Three approaches were described in Section 2. Based on these 
approaches, three types of experiments were conducted to 
confirm the successful performance of gaze tracking using the 
airborne camera simulator. 
Approaches 1 and 2 are the gaze-tracking procedures with a 
known 3-D model. The known 3-D models were generated via 
CAD in Approach 1. They were generated through stereo 
matching measurements in Approach 2. 
Approach 3 is the gaze tracking procedure without a known 
3-D model. Here, a 3-D model was generated via an image 
subtraction algorithm for reference data in the gaze tracking 
procedure. 
4.1.1 Instrument 
The airborne camera simulator developed is shown in Figure 3. 
Figure 4 shows details of the active camera. Figure 5 shows 
acquired images from the active camera. The simulator was 
designed along the lines of the discussion in Section 3.2. 
Figure 3. The airborne Figure 4. Active stereo camera 
stereo camera simulator 
4.1.2 Material 
Diorama building models were prepared for these experiments, 
as shown in Figures 6 and 7. 
Figure 6. Building diorama model (1) 
Figure 7. Building diorama models (2) 
4.2 Prepared 3-D model 
4.2.1 Approach 1: The gaze tracking procedure with a 
known 3-D model (CAD model) 
A set of building diorama models and CAD model shown in 
Figure 8 are used as the material for this experiment. 
Camera 1 Camera 2 Camera 3 
Figure 5. One acquired image shot 
4.2.2 Approach 2: The gaze tracking procedure with a 
known 3-D model (stereo matching procedure) 
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 stereo matching procedure as follows. First, stereo images 
of the building diorama models are captured from 18 directions, 
as shown in Figure 9. Next, 3-D segments of these shots are 
measured from each stereo image. Finally, each 3-D segment is 
merged to generate 3-D data without overlapping 3-D segments. 
The 3-D data generated from this procedure are shown in 
Figure 10.