wierd...I have never got maya to run a compressed mov or avi. Anyways, it will logically be faster to use image sequences as you don't have to run uncompressing algorythms.
carrying on...I am not sure about what you are trying to do...do you mena you want to match a camera or moving camera to live action (camera lineup and match-moving with maya live or Realviz, or Boujou, or 3DEqualizer), or do you mean match a cg object to a moving live action object (rotoscoping)????
There are specific methodologies to both. As for object rotoscoping...depends on what you want to rotoscope as to whether you need to do it by hand, or whether a match-moving software can help you out. It is not a simple thing to get software to do this. The reason is that to get a good object track through automation, the software requires enough depth information to place the object in 3d space in each frame, relative to the camera. This means, if you camera is static, and you have no clear and precise marks or tracking markers to follow on your object, the software will not be able to work it out. You need a minimum of 6 points of info in each frame, plus you need good paralax.
If the camera is moving, it makes the job a little easier, as if you can make a good match-move, there is important paralax data the computer can use to work out where your moving object is. A good trick that most of these match-move softwares will do, is to export the "solve" for a match move in reverse, so the camera is locked off, and the "point cloud" is inheriting the animation. So, if you can't get a good object track using standard methods in the match-move software, you could do that instead..so you would only track 2d points on the moving object for example...so when it does a solve, it will treat the object as static and think that the camera is moving around it. You just reverse it when you go to export with the options for exporting it that way.
Rotoscoping by hand is painful, but often the only realistic option...especially if you have no good tracking markers or lots of motion blurr, or your object is not just moving but also changing shape...like rotoscoping a human arm and hand for example...the muscle deforms the shape, as does of course moving limbs and fingers, and thus totally confusing the software solver and preventing it from working.
There is a lot you should do before you shoot the scene to make this whole process as totally painless as it can be when done correctly...and it can indeed be painless, as opposed to painful...trust me on this one. If you need more info, email me.