Heading Detection From Optic Flow In The Presence Of Human Motion.

A bias in heading estimation is observed if an extraneous object moves so as to obscure the focus of expansion of an optic flow field generated by linear translation. This bias, however, has only been shown for simple, non-articulated objects with no social or behavioural relevance. Contrastingly, objects in the real world are often much more complex and can be comprised of multiple, differently moving elements. Additionally, the movements of particular objects, such as other human actors, may carry information essential to the survival of the observer. In the current study we tested the accuracy of heading estimates in the presence of normal and scrambled point-light walkers. We found that both walkers biased heading estimation when crossing the focus of expansion. This heading bias, however, was significantly larger in the scrambled walker condition than the normal walker condition. Based on these results, we suggest that during translation the visual system is able to use biological form information to efficiently group and segment extraneous points from the optic flow field, thus simplifying the estimation of heading. This has important ramifications for models of heading perception, which generally ignore the characteristics of moving objects and simply treat them as noise. Meeting abstract presented at VSS 2015.