Local computation of angular velocity in rotational visual motion.

Retinal images evolve continuously over time owing to self-motions and to movements in the world. Such an evolving image, also known as optic flow, if arising from natural scenes can be locally decomposed in a Bayesian manner into several elementary components, including translation, expansion, and rotation. To take advantage of this decomposition, the brain has neurons tuned to these types of motions. However, these neurons typically have large receptive fields, often spanning tens of degrees of visual angle. Can neurons such as these compute elementary optic-flow components sufficiently locally to achieve a reasonable decomposition? We show that human discrimination of angular velocity is local. Local discrimination of angular velocity requires an accurate estimation of the center of rotation within the optic-flow field. Inaccuracies in estimating the center of rotation result in a predictable systematic error when one is estimating local angular velocity. Our results show that humans make the predicted errors. We discuss how the brain might estimate the elementary components of the optic flow locally by using large receptive fields.