3D structure perceived from dynamic information: A new theory.

Fulvio Domini
Brain Sciences
Brown University

An important source of information to the three-dimensional 
structure of the environment is the dynamical properties of 
moving images.  The problem of how the visual system uses this 
information has been traditionally studied by assuming that 
the pattern of 2D velocities on the retina is analyzed by the 
visual system by means of a veridical mathematical analysis.  
A new approach that is proposed here is characterized by two 
main hypotheses:  1) the visual system relies on properties of 
the optic flow that are not necessarily sufficient for 
deriving the projected objects and their 3D motion; 2) the 
derivation of the 3D structure and motion is based primarily 
on a heuristic analysis of the optic flow.  I have provided 
evidence for these hypotheses by studying human performance 
in a variety of perceptual tasks that involved judgments of 
3D structure and 3D motion.  These empirical outcomes can be 
predicted by a model that derives local orientation of 3D 
surfaces and their 3D motion from a property of the optic flow 
called deformation.  Since this property is inherently 
ambiguous surface attitude and angular velocity are recovered 
by determining magnitudes of these parameters that most likely 
produce the deformation of the velocity field.