A Neural Model of How the Brain Computes Heading from Optic Flow in Realistic Scenes

Abstract

Animals avoid obstacles and approach goals in novel cluttered environments using visual information, notably optic flow, to compute heading, or direction of travel, with respect to objects in the environment. We present a neural model of how heading is computed that describes interactions among neurons in several visual areas of the primate magnocellular pathway, from retina through V1, MT+, and MSTd. The model produces outputs which are qualitatively and quantitatively similar to human heading estimation data in response to complex natural scenes. The model estimates heading to within 1.5° in random dot or photo-realistically rendered scenes and within 3° in video streams from driving in real-world environments. Simulated rotations of less than 1 degree per second do not affect model performance, but faster simulated rotation rates deteriorate performance, as in humans. The model is part of a larger navigational system that identifies and tracks objects while navigating in cluttered environments.