Are you aware of Peter Kovesi’s web site?
He has developed some Matlab code to compute
the dominant motion transformation between
two successive frames. His code is publicly available
in his web page.

It would be nice if you guys integrated those
matlab routines into matlab as part of the demo
for spatial transformation.

To use matlab spatial transformation routines, one
has to select the control points. But computer vision
researchers have long been developing algorithms to
automate this. They first find features on the two
frames and then they match them. Once they are
matched, a dominant spatial transformation is
computed from matched features using RANSAC.

Hulya