Motion Track Object to drive Pan & Scan?

I am using the VSE to do a picture-in-picure effect. I have two camera angles of a sporting event, and I for the small (inset) image, what I really want to do is tag a small window (original video is 1080p, so I want to use, say, a 640x480 subframe rather than scaling) from the second camera to certain objects within the strip.

I have approximated this effect so far by using a Transform, cropping, and keyframing the X and Y position. The problem is, the second camera is hand-held and very jittery.

So, what I want to be able to do is to create that small subframe using a view that is attached to a motion track of the objects within the image. I have already built the motion trackers; what I am not sure how to do is to attach a 640x480 viewport to it in order to generate a new strip for the overall sequence.

Any ideas? Do I need to delve into the compositor? Into separate scenes added as scene strips?

Thanks for any help!


Solution here: