A neural network model for detecting a planar surface spatially from the optical flow in area MST of the visual cortex

<jats:title>Abstract</jats:title><jats:p>The psychologist Gibson found some 50 years ago that motion stereo, which is indispensable in walking or driving, is realized via the optical flow on retina caused by motion. The stereo contains detection of the direction of self movement and spatial perception of the plane (its three‐dimensional orientation and depth) composing the environment. With the development of neurophysiology, it has been found that there exist cells that detect the direction of self movement from an optical flow in area MST (medial superior temporal area) of the visual cortex, acting as the central system for motion perception. The cell that spatially perceives a plane on the basis of the optical flow has not been found, despite the psychological findings of Gibson. Neither has there been a report on such a model. This paper reports an algorithm that detects three types of plane parameter (i.e., three‐dimensional orientation of the plane, time‐to‐contact to the plane, and shortest distance to the plane) by integrating the optical flow. The time‐to‐contact and orientation are detected by using the cross‐ratio transformation and polar transformation. The shortest distance and orientation are detected by using the small‐circle transformation. On the basis of these proposed algorithms and a previously reported neural network model that detects local image motion in area MT of the visual cortex (Vision Research 1996, 1999), a neural network from retina to area MST through area MT is modeled, which detects the three parameters of a plane. The neural network model (composed of 0.16 × 10<jats:sup>8</jats:sup> cells and 4.6 × 10<jats:sup>8</jats:sup> connections among these) is constructed on a computer, and it is shown that the time‐to‐contact and orientation are correctly detected. © 2003 Wiley Periodicals, Inc. Syst Comp Jpn, 34(4): 46–59, 2003; Published online in Wiley InterScience (<jats:ext-link xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="http://www.interscience.wiley.com">www.interscience.wiley.com</jats:ext-link>). DOI 10.1002/scj.1200</jats:p>