Adverti horiz upsell
Generating A Matched Reveal Source For Painting
Generating A Matched Reveal Source For Painting
mplec, added 2005-09-17 22:21:16 UTC 35,222 views  Rating:
(1 rating)
Page 2 of 5

Track the Plate


Since I'm going to be repositioning one image to match another, I know I'll need to get tracking information on the plate. A full tracking tutorial is beyond the scope of this article, but there are a few aspects of this shot worth noting. A look at the footage for this section of the shot shows that the wire to be removed is generally passing over the window in back.

click for larger version

Given the extent of the moving camera throughout the shot, a 2 point matchmove would not account for all the changes in the appearance of the window area. A four-point track will allow me to pin edges of the window in place to get the most accurate match possible in that window area. This may cause the image to stretch and match poorly farther out towards the edge of frame, but only the areas within and near the window are required for the paint-through. As long as that lines up well the rest doesn't matter; it won't be used.

The tracker's default (and most reliable) mode is luminance. Here I've attached a Reorder (to "lll", putting luminance in each channel) and then an Expand to guide me as I position the Tracker's feature areas.

click for larger version

Sometimes it's a bit hard to tell what might be good track points; our eyes and brain are not processing the image in the same way that the Tracker's logic will. We may recognize certain groups of pixels as unique features even though they don't look so well-defined to Shake. Before settling on tracking points it can be helpful to take a look at the image's luminance and slide the range around a bit to see what pops out.

click for larger version

While it is at times helpful to preprocess the image a bit before tracking, particularly to remove grain that might throw off the correlation computations, I'll first try tracking directly on the original source footage. The Tracker is built to expect regular images and does some normalization itself to best compute the correlation. If it seems to be having trouble I'll spend some time adjusting the image to help it out. With this footage it doesn't turn out to be necessary.

In the simplest case this tracking data would now be used to lock another piece of footage to the original; for example, some prepared footage placed into an on-set television screen. Another common situation involves a piece of footage that's moving and needs to be tracked into position on the original moving plate. This shot is a variation on that second case, but since the second piece of moving footage is the same as the first I've already got the tracking data. The trick is setting up so that I can line up any arbitrary frame of it with any other, rather than just lining up one running sequence frame by frame with the other.