A MODEL OF THE COMBINATION OF OPTIC FLOW AND EXTRARETINAL EYE-MOVEMENT SIGNALS IN PRIMATE EXTRASTRIATE VISUAL-CORTEX - NEURAL MODEL OF SELF-MOTION FROM OPTIC FLOW AND EXTRARETINAL CUES
M. Lappe, A MODEL OF THE COMBINATION OF OPTIC FLOW AND EXTRARETINAL EYE-MOVEMENT SIGNALS IN PRIMATE EXTRASTRIATE VISUAL-CORTEX - NEURAL MODEL OF SELF-MOTION FROM OPTIC FLOW AND EXTRARETINAL CUES, Neural networks, 11(3), 1998, pp. 397-414
The determination of the direction of heading from optic flow is a com
plicated task. To solve it the visual system complements the optic flo
w by non-visual information about the occurrence of eye movements. Psy
chophysical studies have shown that the need for this combination depe
nds on the structure the visual scene. In a depth-rich visual environm
ent motion parallax can be exploited to differentiate self-translation
from eye rotation. In the absence of motion parallax, i.e. in the cas
e of movement towards a frontoparallel plane, extraretinal signals are
necessary for correct heading perception (Warren and Hannon, 1990). L
appe and Rauschecker (1993b) have proposed a model of visual heading d
etection that reproduces many of the psychophysical findings in the ab
sence of extraretinal input and links them to properties of single neu
rons in the primate visual cortex. The present work proposes a neural
network model that integrates extraretinal signals into this network.
The model is compared with psychophysical and neurophysiological data
rom experiments in human and non-human primates. The combined visual/e
xtraretinal model reproduces human behavior in the case of movement to
wards a frontoparallel plane. Single model neurons exhibit several sim
ilarities to neurons from the medial superior temporal (MST) area of t
he macaque monkey. Similar to MST cells (Erickson and Thier, 1991) the
y differentiate between self-induced visual motion that results from e
ye movements in a stationary environment, and real motion in the envir
onment. The model predicts that this differentiation can also be achie
ved visually, i.e. without extraretinal input. Other simulations follo
wed experiments by Bradley et al. (1996), in which flow fields were pr
esented that simulated observer translation towards a frontoparallel p
lane plus an eye rotation. Similar to MST cells, model neurons shift t
heir preference for the focus of expansion along the direction of the
eye movement when extraretinal input is not available. They respond to
the retinal location of the focus of expansion which is shifted by th
e eye movement. In the presence of extraretinal input the preference f
or the focus of expansion is largely invariant to eye movements and ti
ed to the location of the focus of expansion with regard to the visual
scene. The model proposes that extraretinal compensation for eye move
ments need not be perfect in single neurons to achieve accurate headin
g detection. It thereby shows that the incomplete compensation found i
n most MST neurons is sufficient to explain the psychophysical data. (
C) 1998 Elsevier Science Ltd. All rights reserved.