A Computational Model of the Perceived Velocity of Moving Plaids
Local motion detection mechanisms generally lead to one component of the optic flow becoming indeterminate. One way to solve this ‘aperture problem’ is to compute the optic flow which minimises some smoothing constraint. With iterative schemes the computed velocity array is suboptimal relative to the constraint until the process has converged. Under the original assumption that the iteration rate is sufficiently low to allow the perception of suboptimal flows at short stimulus durations, iterative gradient models give an accurate description of biases in the perception of tilted line velocity. We examine whether this approach can be applied to moving sinusoidal plaids. Our simulations are in agreement with a number of psychophysical results on both speed and direction perception. In particular we show that the effect of stimulus duration on the perceived direction of type II plaids [Yo and Wilson, 1992 Vision Research32(1)] can be accounted for without recourse to second-order mechanisms. The effects of contrast and component directions on the evolution rate of this bias are well reproduced. The model also successfully describes the effect of spatial frequency, and data obtained with gratings. These results suggest that iterative gradient schemes can model the dynamics of interactions between local velocity detectors, as revealed by psychophysical experiments with lines and plaids.