Abstract
Several studies have revealed that human heading perception based on optic flow is biased when independently moving objects (IMOs) cross or approach the observer’s future path. However, these biases are surprisingly weak (~2°) and perceived heading does not seem to abruptly shift at the moment that a moving object crosses the observer’s future path. While previous studies have focused on biases, it is equally important to understand how heading perception is as robust and stable as it is. Such robustness and stability is surprising given that IMOs often occupy large portions of the visual field and occlude the focus of expansion (FoE). Why isn’t heading perception, which is based on neurons tuned to radial expansion, biased by more than several degrees or abruptly shift when an object crosses the future path? Indeed, our simulations of existing models (differential motion and center-weighted template models) yield heading estimates that are far more erratic and unstable than human judgments. We present a dynamical model of primate visual areas V1, MT, and MSTd based on that of Layton, Mingolla, and Browning (2012) that explains how the visual system reliably estimates heading during navigation through dynamic environments. Unlike existing models, competitive dynamics between units in MSTd stabilize the model’s heading estimate over time, even when an IMO crosses the future path. Soft winner-take-all dynamics enhance units that code a heading direction consistent with the time history and suppress responses to transient changes to the optic flow field. The model explains the surprising bias that occurs when an object disoccludes the future path, although the FoE is visible in the flow field (Layton & Fajen, Submitted to JoV). Our findings support competitive temporal dynamics as a crucial mechanism underlying the robustness of perception of heading.
Meeting abstract presented at VSS 2015