A recurrent Bayesian model of dynamic motion integration for smooth pursuit

Abstract

The quality of the estimate of an object's global motion, over time is not only affected by the noise in motion information but also by the spatial limitation of the local motion analyzers (aperture problem). Perceptual and oculomotor data demonstrate that during the initial stages of the motion information processing, 1D motion cues related to the objects edges have a dominating influence over the estimate of the objects global motion. However, during the later stages, 2D motion cues related to terminators (edge-endings) progressively take over leading to a final correct estimate of the objects global motion. Here, we propose a recursive extension to the Bayesian framework to describe the dynamic integration of 1D and 2D motion information. In the recurrent Bayesian framework, the prior defined in the velocity space is combined with the two independent measurement likelihood functions (Likelihood functions representing edge-related and terminator-related information) to obtain the posterior. The prior is updated with the posterior at the end of each iteration step. The recurrent Bayesian network is cascaded with a first order filter to mimic the oculomotor dynamics in the final output of the model. This oculomotor dynamics was tuned with single blobs moving in 8 different directions. The model parameters were fitted to human smooth pursuit recordings for different stimulus parameters (speed, contrast) across three subjects. The model results indicate that for a given velocity with increase in contrast, the latency decreases and for a given contrast with increase in velocity, the acceleration increases similar to what is being observed in smooth pursuit recordings. Also, The latency for a tilted line is shorter compared to the latency for the blob.

Footnotes

 CODDE project (EU Marie Curie ITN), CNRS.