The human visual system must operate in a rapidly changing environment, as objects, eyes and obsei'ver are continually moving. This fact must, to a great extent, determine how the system analyses its retinal input, We argue that perception should be regarded as a dynamic process in which patterns of neural activity are developed and change in ways which reflect changes in the visual field.
A model is described to suggest how the visual system may keep track of perceived objects as their images move on the retina. We postulate that at some level of the visual system, the position of these objects relative to the observer is represented by a pattern of neural activity. Such a pattern of activity must move as a unit as the object it represents moves: the pattern cannot be continually regenerated during motion.
In the model we propose, information is represented by activity in two-dimensional, homogeneous layers of neuron-like elements. A number of these layers are arranged in parallel, and activity within separate layers represents object position and the components of retinal image velocity due to object and observer motion. Object velocity, which cannot be directly sensed at the level of the retina, is isolated and represented analogically. This representational isolation of object-and observer-related velocity allows us to explain several illusions of motion perception, including induced motion and the "waterfall effect".