Dynamic inference for motion tracking
Abstract
When the visual information about an object’s motion differs at the local level, the visuomotor system needs to integrate information across time to solve this ambiguity and converge to the final motion solution. For an oblique line moving horizontally, edge-related motion cues differ from terminator-related information, the latter being coherent with the line’s global motion. We have previously shown that ocular tracking of this kind of stimuli is transiently biased toward the edge-orthogonal direction, before converging to the global motion direction. Here, we model the dynamic convergence to the global-motion solution as a recursive update of inferential knowledge in the velocity space. We assume that motion estimation is based on a prior distribution and two independent likelihood functions representing edge-related and terminator-related information. Importantly, the shape of the Bayesian functions is constrained by smooth-pursuit eye-movement data. Model predictions about the dynamic convergence to the correct motion solution are compared to human smooth-pursuit recordings when varying different stimulus parameters (speed, contrast).
Laurent U Perrinet
Researcher in Computational Neuroscience
My research interests include Machine Learning and computational neuroscience applied to Vision.