It depends what you want to do as what you have described could mean many things. If you want to have a movement in the camera frame resulting in a movement to on the screen you could write a frame difference algorithm and use this to drive motion.

A more advanced technique would be to use a edge detection or blob detection library (v3ga's or JMyron for instance) can do this. The main drawback to doing it in Java is java. It really isn't the best language to be doing something so intensive and you need cutting edge equipment to make it work in java.

There are other packages out there that may be slightly better at this like Eyesweb, ARToolKit and MAX/Jitter but they all have their pluses and minuses.

Some tips for you to help performance are the smaller the camera captured image the less processing it's going to take. If you can work out a way so that the screen object moves from an interpolated result from the camera image it may make things easier if you only update the camera intermittently.