The theory of affine transfer presented
in [15, 5]
is a method of choosing a fixation point on a tracked object,
so that image-plane errors of the fixation point from the centre of
the image(s) may be fed back to the motors controlling the camera(s).
In this way a desired point on the object may be maintained at or near
the centre of the image for as long as the object is tracked.
The algorithm involves constructing a coordinate frame centred on the
object, choosing a fixation point in that coordinate frame,
and transferring the point to new images as they arrive.
In the context of the 2D affine stereo reconstruction algorithm described
above, this corresponds to selecting
a point 
in the space of the structure vectors as the fixation point.
Given the recursively computed scene reconstruction, the transfer
of the fixation point into the new stereopair with computed motion parameters
and 
is simply
The transferred positions 
, 
are now converted
via triangulation to 3D world coordinates to be interpreted as
a range measurement.
It seems sensible to choose the centroid of the structure vectors
computed from the initial batch computation of the 2D reconstruction for
, and this is currently what we do.