5 Architectures
The goal of an expanded architecture is not only to
improve performance of current systems and application
but to allow computer vision to take on the
aspects of intelligence which human vision systems
exhibit. Humans use their vision system and other
parts of their brain to gain what is termed Situational
Awareness, that aspect of intelligence that allows humans
to understand their environment and interact
within it. Through the use of intelligent systems, we
can incorporate a computer vision system for modeling
the system state. By combining these systems, we
have a system capable of system and environmental
situational awareness. In order to realize the potential
for computer vision, more complex architectures
are needed. I will examine four architectures that
demonstrate how combined AI systems can be used to
increase the performance of computer vision systems.
These architectures are being used in my research ancl
in that of a number of other researchers with grcat success. The first approach is a pipeline architecture
where the processing components are grouped based
on their functions in the overall computer vision system,
see figure 1. The first component is the data
acquisition. In this component, time sequencing of
inputs and data fusion functions will be addressed.
In the second, component image pre-processing will
take place. These functions are similar to the transfer
functions in the human vision system, which include
edge detectors and filtering functions. The third component
is used to look at the image and detect or
identify parts of the image. The fourth component
is the implementation of intelligent decisions about
what the image is and its meaning. The last component
of the system will close the loop for the system
with feedback and control.