Page 29 - Dynamic Vision for Perception and Control of Motion
P. 29
1.5 What Type of Vision System Is Most Adequate? 13
To detect objects in special areas of interest far away, a camera with a third fo-
cal length (again with a factor of 3 to 4 relative to the mild telelens), and the field
of view within that of the mild telecamera should be added (see Figure 1.4). This
camera may be chosen to be especially light-sensitive; black-and-white images
may be sufficient to limit the data rate. The focal length ratio of 4 does have the
advantage that the coarser image represents the same scene at a resolution corre-
sponding to the second pyramidal stage of the finer one.
This type of sensor combination is ideally suited for active viewing direction
control: the coarse resolution, large simultaneous field of view allows discovering
objects of possible interest in a wide area, and a viewing direction change will
bring this object into the center of the images with higher resolution. Compared to
a camera arrangement with maximal resolution in the same entire field of view, the
solution shown has only 2 to 4 % the data rate. It achieves this in exchange for the
need of fast viewing direction control and at the expense of delay times required to
perform these gaze changes. Figure 1.5 gives an impression of the fields of view of
this trifocal camera arrangement.
Figure 1.5. Fields of view of trifocal camera arrangement. Bottom: Two divergent wide
angle cameras; top left: mild tele camera, top right: strong tele-camera. Dashed white
lines show enlarged sections
The lower two wide-angle images have a central region of overlap marked by
vertical white lines. To the left, the full road junction is imaged with one car com-
ing out of the crossroad and another one just turning into the crossroad; the rear of
this vehicle and the vehicle directly in front can be seen in the upper left image of
the mild telecamera. This even allows trinocular stereo interpretation. The region
marked in white in this mild teleimage is shown in the upper right as a full image