If you are looking for translation of the image, not the ROI, the best thing to use is probably pattern matching. It would only work for a static object, but you could train an area and monitor its coordinates using pattern matching.
Bruce
Bruce Ammons
Ammons Engineering