Modelling Fly Motion Vision - Computational Neuroscience: A Comprehensive Approach - page 393

study of motion vision. Motion vision has thus become a classical problem in com-
putational neuroscience, which many laboratories around the world have embarked
on. In the following I will give an overview of what is known about the computations
underlying motion vision in the fly, where a lot of experimental results are available
(for review see: [13, 36]) and where modelling efforts have reached a rather detailed
biophysical level at many processing steps.
This chapter summarizes our current understanding of fly motion vision with an
emphasis on modelling rather than on the large set of available experimental data.
After giving an overview of the fly motion vision system, the next part of the chapter
introduces the correlation-type of motion detector, a model for local motion detection
that has been successfully applied to explain many features of motion vision, not only
in flies but also in higher vertebrates including man. This is followed by an outline
of how local motion signals become spatially processed by large-field neurons of
the lobula plate in order to extract meaningful signals for visual course control. In
a final section, the article will discuss in what directions current research efforts are
pointing to fill in the missing pieces.
The processing of visual motion starts in the eye. In flies, like in most inverte-
brates, this structure is built from many single elements called facets or ommatidia.
Each ommatidium possesses its own little lens and its own set of photoreceptors. The
latter send their axons into a part of the brain exclusively devoted to image process-
ing called the visual ganglia . Within these ganglia, images become processed by an
array of local motion detectors ( Figure 14.1 , top). Such motion detectors are thought
to exist for horizontal as well as for vertical image motion and to cover the whole
visual field of the animal.
In the next processing step the output of such local motion detectors become spa-
tially integrated by various large field elements. Anatomically, this happens on the
dendrites of tangential cells located in the posterior part of the third visual ganglion,
called the lobula plate . There exists a limited set of such tangential cells that can
be grouped according to their preferred direction of image motion (Figure 14.1, bot-
tom): some of the cells respond preferentially to horizontal image motion from front
to back (e.g., the three HS-cells, i.e., HSN, HSE and HSS, both CH-cells, i.e., dCH
and vCH), others to horizontal image motion in the opposite direction (H1 and H2),
others respond selectively to vertical image motion from top to bottom (the VS-cells
VS1, VS2, VS3 ) etc. All in all, there are only about 60 such neurons per hemisphere
in the blowfly Calliphora that collectively cover the whole visual field of the animal.
However, these neurons do not integrate the output signals of local motion de-
tectors independently but interact with each other. Specific connections have been
determined between tangential neurons of the left and the right lobula plate as well as
between neurons within one lobula plate (Figure 14.1, colored lines). These connec-
tions tune many tangential cells responsive to specific motion signals in front of both
eyes, and others that are selectively responsive to motion of small moving objects
or relative motion. Tangential cells have been shown to synapse onto descending
neurons (e.g., [86]) which connect either to the flight motor in the thoracic ganglion
of the animals controlling the various flight maneuvers, or to specific neck muscles
controlling head movements (not shown).