Biomimetic Vision Sensors - Engineered Biomimicry - page 23

brief. The literature provides a rich heritage of
this extensive body of work.
In the early vision processes, there is consider-
able visual information to be gleaned from the
photoreceptors and their interaction with the
first several cellular layers. There is enough evi-
dence in the literature that fairly significant
vision processes may take place early in the
vision system (prior to the brain). For example,
several researchers studying single-unit recor-
dings in early vision processes of various orga-
nisms found visual neurons that had sensitivity to
moving images [44-46] . Specifically, they noted
the impulse rates of specific cells were modified
by changes in the visual stimulus direction. Marr
also posited that primitive forms of object recog-
nition may take place on the retina. He noted that
single neurons may perform more complex pro-
cessing tasks than had been previously thought,
including the ability to detect pattern elements
and to discriminate the depth of objects [2] . Marr
based his conclusions on the experiments con-
ducted on frogs by Barlow, who observed that
the frog's selective stimulation of neurons at the
retinal layer served as bug detectors. That is, the
neurons were providing a primitive form of
object recognition on the retina [2, 44] .

-

FIGURE 1.15 Convexity cell for extracting optical flow.
Adapted from Ref. 47 .

peripheral field of view appear to move faster.
Sensors that are specifically sensitive to optical
flow have been designed [48, 49] .
The elegance of the Nakayama-Loomis
theory is the ability to compute optical flow glob-
ally by employing local, simple neuronal net-
works, as depicted in Figure 1.15 . Work by
O'Carroll on dragonflies further noted that these
and other related insects possess neurons that are
tuned for detecting specific pattern features such
as oriented line edges and moving spots [50] .

1.3.3.1 Optical Flow
Additional work by Nakayama and Loomis
further supported sophisticated processing at
the retinal level. These researchers postulated
that center-surround motion-detection neurons
exist and are directionally sensitive to different
stimuli [47] . Several motion-detector cells, with
different orientation sensitivities, feed a higher-
order convexity cell to produce optical flow.
Optical flow. is the apparent motion of surfaces
or objects in a scene, resulting from the motion
difference between the observer and a scene. An
example of optical flow, as discussed in Chapter
9 by Chahl and Mizutani, is evident when driv-
ing a vehicle: when we looking straight ahead,
objects appear to be stationary; as we shift
our gaze gradually to the side, objects in the

1.3.3.2 Motion Processing
The literature includes a considerable body of
work that has been accomplished to describe
vision and motion processing at higher-order
process regions of the vision system. There are
four broad categories of motion-processing
models [43, 51] :

• Differential-orgradient-basedmodels
employing first and second derivatives to
determine velocity;
• Region-orfeature-basedmatchingto
determine movement between adjacent
temporal image scenes (frames);