There is
this thread about it already.
Briefly, I think the only thing of this sort that could actually be done is "shoot when no motion is detected".
(Because shooting many short exposures and then merging them is prohibitively expensive in terms of memory speed... and the algorithm for alignment isn't trivial, either.)
There is already
a little hack in the form of a script that uses motion detection "in reverse", although it's of course far from optimal, and to have any chance of being used for image stabilization purposes, you'd definitely have to put it in the C code of CHDK.
