Page 87 - Dynamic Vision for Perception and Control of Motion

3.3 Perceptual Capabilities 71

bilities explicitly on the abstract level in the capability network, the process “cen-
tral decision” (CD, upper right) can parameterize and use it in a flexible way. The
capability OVB depends on the availability of the complex skill saccade and
smooth pursuit. It possesses as parameters the maximal number of saccades, the
planning horizon, eventually, a constant angular position for one of the platform
axes, and the potential for initiating a new plan for gaze direction control. The de-
mand of attention and the combination of regions of attention for certain objects to
be observed are communicated by the specialists for recognizing and tracking these
objects (see Chapters 13 and 14). Complex patterns in visual perception may
emerge this way depending on the priorities set in the system. The second sche-
matic capability beside OVB on this level is 3-D search. This allows scanning a
certain area in the environment of the vehicle in the real world by sequences of
saccades and scans. The scans are performed with constant angular speeds so that
image evaluation is possible; saccades may be interspersed so that the scanning di-
rection is always the same. Scan rate may depend on the distance viewed.
Central decision (CD), the process in charge of achieving the goals of the mis-
sion, has contact only with BDGA, not to the lower levels directly. This modulari-
zation alleviates system development and naturally leads to multiple scales (coarse-
to-fine differentiation). It should have become clear that this scheme allows charac-
terizing vision systems to a relatively fine degree. Compact representation schemes
for a wide variety of vision systems are possible and left open for future develop-
ments. The concept has been designed to be flexible and easily expandable.

3.3.2.4 Feature Extraction Capabilities

Beside the capabilities of gaze control, the capabilities of visual feature extraction
characterize the performance level achievable by a subject. Thresholds in percep-
tion of edge and corner features are as important as recognizing shades of gray val-
ues or colors in a stable way. Recognizing shapes originating from boundaries of
homogeneous image areas or from smooth or connected boundary sections allows
inferences for hypothesis generation of objects or subjects seen, especially when
continuity conditions over time can be discovered and also tracked. This will be
one of the major topics of this book.
Biological vision systems have developed a high standard in recognizing tex-
tures, even two different ones simultaneously, as when one surface moves behind a
partially obscuring other object (for example, an animal behind a tree or bush). The
state of development of processing power of computers does not yet allow this in
technical systems. In biological evolution, in certain situations like a predator ap-
proaching prey, maybe only those prey animals had a chance to survive which
were able to solve this problem sufficiently well. For many applications of techni-
cal systems, this high level of visual capabilities is probably not necessary.