how can I trace a ROI with fixed size on multiple images ?

I am trying to register images from a stack. they are frames taken from a video in which an object is moving. I tried to use the

routine described here (https://ch.mathworks.com/help/images/control-point-registration.html).

Everything works fine however since I am using a drawpolygon function to detect the control points, each image gets scaled differently in the output (rotation and traslation are correct). Is there a way to use the control points to apply a rigid transformation (without imwarp maybe ?).

I thought about using imrotate plus a simple transformation of the cartesian coordinates...doing some averaging (for instance the angles of the ROI segments belonging to the fixed and to the moved image can be calculated, averaged and then the average angle used to rotate the image... same would apply to the translation).

Is there a more streamlined/easy way to achieve this ? the main issue is that the ROI is customizable each time... instead, if it was fixed... it would be perfect... but I am not sure how one could design a ROI with fixed dimensions but customizable rotation and translation. it sounds even more difficult than the solution mentioned above.

I would appreciate any advice