Learning Structure from Motion: How to Represent Two-Valued Functions

Abstract

The optic flow is the vector field formed by the projection of the 3D-motion in the environment on the image plane of the observer. The optic flow vector o at image location r is thus a function of the scene depth z _r and the relative scene motion m _r = (t, ω) _r which can be written as

$$ o_r = C(r,{m_r},{z_r}) $$

where C is some non-linear function which only depends on the camera mapping.

Real World Computing Partnership

Dutch Foundation for Neural Networks