How to Fake a UFO Landing
First, head over to www.digilab.uni-hannover.de/download.html, and download a package appropriate to your system. Note that there are no source packages—Voodoo may be freeware, but it is not, and probably never will be, open source. So, grab the binary that is convenient for you. Note that there are no x86_64 binaries available. If you have a 64-bit system, just grab the x86 package—it doesn't depend on any 32-bit libs to work, and it won't choke on execution.
Pop open a command window, and use tar -xvzf to open the archive. Next, move into the resulting voodoo-versionnumber directory, then a further level deep into the /bin directory, and run ./voodoo.
Anyone familiar with old Roger Corman movies will realize that bloodletting is an essential step in working good voodoo. In this case, it's your video that needs to be bashed into tiny bits. Voodoo will not chew through video, it works only on still image sequences.
A quick ffmpeg call will give you the image sequence you require:
ffmpeg -i videofilename.avi -f image2 %03d.png
Once done, run Voodoo.
The interface, at first glance, is simple—two pull-down menus and a flipbook player. That simplicity proves quite illusory as you begin to delve into it. Camera tracking is complex, and the toolbox here is extensive, but the nature of the task means that you can learn gradually, and very little work will get you a good initial track.
To start, go to the File menu and select load→image sequence, and load up the image sequence you just created. Be sure to set movement type and interlace settings, or your track will not come out properly. Play the clip through once with the flipbook to make sure there aren't any obvious errors.
Now, you need to load camera settings (File→load→initial camera). This is vital if you want the track to work properly, but it's also very difficult to get right if you weren't keeping notes on the set for focal length, aspect ratio, film back and (less important) skew angles. If you didn't keep proper notes, enter your best guess and go from there, you always can tweak it later.
The work flow from here is pretty simple. Play through the clip to make sure the whole sequence loaded properly, then press track. The computer will select a few dozen track points and follow them through the duration of the clip. Depending on the complexity and length of the clip, this process can run anywhere from a few seconds to a few hours.
Once the track is done, play it through again, watching particularly the motion path of the different points. If you can't see any drift, you're golden—you can skip ahead to the export step. If the track is lacking, there are a number of ways to tweak it. You can refine by adjusting the tracking algorithms in the View→Controls menu, and rerun the track, selecting refine instead of discard in the dialog that presents itself to augment the track you've already created. You can do much the same through adjusting the camera settings, although if you do this, you'll be better off running the track from scratch.
A number of other refinement tools are also available. You can pull up the modeling box (View→Modeling Tools) and use it to add track masks and 3-D primitives to help you spot drift, and it (along with the Fpoint track editor) lets you delete, change or add new track points manually, so you can direct the tracker to watch the right things and make it ignore the wrong ones, such as people or cars in the foreground. Once done, run the track again, again selecting refine rather than discard.
You can watch the reconstructed camera motion, and manipulate it to a certain extent, in the 3-D viewer window, available through the View menu.
When you have a track you find satisfactory, go to File→Save, and pick your export format. Be sure to export all the Fpoints—having them helps if you're going to need to do any complex interaction, as they will help you guide where you put alpha masks and such—like if you chose to do some of your masking in your 3-D program.