The neural networks that provide stereoscopic vision

Abstract

Stereoscopic vision is the capability of perceiving the three-dimensional shape of objects and of determining their distance and relative position in space, based on the differences (disparities) of the images of an object on the two retinas. The goal of this paper was to consider how the neural networks that analyze the disparity in the primary visual cortex are organized. Information is presented that is evidence of the optimal nature of the organization of the visual system for forming neural structures that produce convergence of the inputs of the different eyes. A map of the placement of monocular projections in the primary visual cortex is constructed for binocularly visible objects of space, on the basis of which binocular neurons that are selective for disparity are first formed. An analysis of the map shows that the binocular neurons formed by simple convergence of monocular cells can provide adjustment to the positional depth relative to the bifixation point—i.e., they can encode the absolute disparity—and this agrees with the available results of neurophysiological studies. Experimental data are given that show that the problem of the correspondence of the two images of the object is not solved in the primary visual cortex. The signals of the binocular neurons of this field in this case are used to control the vergent movements of the eyes.