tracking to automate resizing and position ?

The camera setup is just simple math. If you shot the footage with a pretty normal lens I'd pick the 50mm preset, depth of field off and two node camera:

Note the zoom value. That is the distance this camera will be placed from comp center when it is created.

Now set up your composition. In this example, the sample footage layers (Image 1, 2, 3 & 4) are all 2X larger than the composition. Make all layers 3D, reveal the zoom value of the camera by pressing A twice, select it and copy it. Next, select footage layer 2, 3 and 4, reveal the position property by pressing P, double-click on the Z value and paste the zoom value to the Z position. Now go to footage layer 3, double-click the Z value, press the right arrow key to move to the end of the z value and type *2 to multiply the value by 2. Repeat for Image 4 except type *3. This will line your images up in z space so that the images will be at an effective 100% scale when the camera moves, in this case, 2666.6667 pixels from the next layer. Add a null and parent all of the Image layers to the null then animate the position of the null. The setup would look like this:

As long as your images are properly sized and lined up so that when the camera is at the effective 100% scale position you can just cut set the out point for each successive layer at the match point and the transition will be seamless. You can use a blend mode to precisely line up each successive image. When you get the images to match up return the blend mode to Normal. The comp would look like this before you changed the blend mode back to normal so you could see that the images are lined up.

Because you always cut at an effective 100% scale and the images perfectly line up at that point you should only need 2 position keyframes and an easy ease applied to make a nicely timed zoom in on the shot. Just back the null up until the first image fills the frame to start.

It's all pretty simple once you have the footage prepared. Getting the layers properly lined up in Z space so the scale is accurate is just simple arithmetic and lining them up in X and Y is simplified by temporarily using blend modes.

It's possible you could use the Warp Stabilizer to help unify these shots, but depending on how different they are, you may just need to do it manually.

The "Difference" blending mode is excellent for this kind of stuff. You could place each shot atop the previous one, set to Difference, and you'll be able to see the adjustments needed to make it match perfectly. It's likely even an automated tool will only get you close, and you may need to take this approach to make it perfect.