Reference frames for disparity and flow

I have made the argument, over a series of papers, that binocular disparities are used by the visual system to recover the relief of points relative to a locally-defined surface. This has the consequence that the measure of depth that is stored is invariant to a large extent across head movements. I have shown that disparity relative to a surface is a better predictor of stereoacuity thresholds than relative disparity. In this sense, current models of the horopter (region of highest stereoacuity) need to be re-evaluated.

• Glennerster, McKee and Birch (2002)
• Glennerster and McKee (2004)
• Petrov and Glennerster (2004)
• Petrov and Glennerster (2006)

A similar system for defining structure relative to a local plane probably applies for structure from motion, too, along the lines suggested by Koenderink and van Doorn (1991):

• Hogervorst, Glennerster and Eagle (2003)

The surface-based system for defining location can be part of a hierarchical structure in which the location of the surface is defined relative to others. This would be a very different way of defining the location of a point compared to recovering the 3D layout of the scene and the observer's location within it, as most current theories (and computer vision systems) assume. Aspects of this model are explored in:

• Glennerster, Hansard and Fitzgibbon (2001)

This paper also discusses how direction of heading can be computed in a hierarchical way, with progressive refinement of the estimate as more information is added. (See also an introduction to the problem in Glennerster (2002)).