ABSTRACT
This paper presents a simulation of neural networks for motion interpretation of trochoidal paths generated by points placed on a rotating disc translating in the frontoparallel plane of the visual field. We posit this motion interpretation process involves two mechanisms: first the extraction of relative rotational motion with respect to a moving point and second the presence of an attention mechanism that tracks the common motion. Specifically, the paper demonstrates simulation results for rotational motion extraction from trochoidal paths. We construct our neural network according to the directional and speed sensitivity profiles observed in neurophysiological data in the first stage of the network. The layer of neural network which is sensitive to the intermeditate spatiotemporal pattern of rotational motion is then assembled. Lastly, we show that without an attention mechanism the perception is insufficient to elicit a translating wheel perception. However, with an attention mechanism to track the center of the rotation, the rotating wheel is perceived readily.
