Purchase this article with an account.
Bo Hu, David Knill, Christopher Brown; Modeling dynamic re-weighting in visual cue integration. Journal of Vision 2005;5(8):402. https://doi.org/10.1167/5.8.402.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Much of the existing data on sensory cue combination is consistent with a linear integration model in which cue weights vary in inverse proportion to the uncertainties of the cues. Research has focused on static cue integration problems, in which cue weights are assumed to remain constant over time. In many scenarios, however, information from one cue can accrue over time to help disambiguate the information provided by another cue, a form of cue promotion. In particular, information from one cue can indirectly disambiguate hidden scene variables that determine the values of other cues. Consider the case of cast shadows. Cast shadows provide information about the relative 3D position of an object and a background surface, but this information is only as reliable as the observer's knowledge of the light source direction. In a dynamic scene, other cues such as stereo, by specifying the relative depth of the object and surface, can indirectly disambiguate the light source direction. In general, one would expect the uncertainty about the light source to decrease over time. This dynamic changing of uncertainties will manifest in the apparent changing of the relative weights of the two cues over time. Similar arguments apply to dynamic cues such as changing size, with object size as the hidden variable. We use Dynamic Bayes Networks to model these scenarios and make connections of our model to the weak fusion model and other cue integration phenomenon such as perceptual-explaining-away. We approximate the optimal non-linear cue integration model using Extended Kalman Filters. The results show that, as predicted, the relative influence of an indirect cue like cast shadows increases with increased exposure time in the presence of direct depth cues like stereo. The model makes strong psychophysical predictions about how the apparent weights of different cues should change over time in dynamic scenes.
This PDF is available to Subscribers Only