In part 1 we discussed the manual 3D tracker. Using the automatic tracker is not just a decision to be based on ease of use. In the previous example the automatic tracker has problems as many of the foreground points are on the car – which of course is not wanted as part of the solution of the track – since it is moving. To track this shot, we needed to select and track points behind the car or well clear to the camera right. Tracks automatically tracking the car are unwelcome and so really most the points tracked ended up being in the top 50% of the screen only.

Flame’s manual tracker works well on some complex problems – but on others it is extremely awkward, primarily if you need to provide many trackers since almost no point is on screen for very long before being blocked or leaving frame. Flame’s manual tracker actually has poor control of which trackers are used. In the shot shown below – the camera swings around , while pulling back with a bar full of people. Any tracker is quickly lost, so naturally one manually tracks multiple trackers that overlap, but determining how many is needed is pure guess work, and to add to the complexity, if the co-ordinates or relationships are changed using the manual 3D tracking option, the software chooses a different sub set of trackers to use and this can produce vastly different results.

Using the automatic tracker is of course extremely simple. One simply selects the menu and presses Autotrack. The automatic tracker acts as a background task so not only can you continue to build your action setup, but you can leave action and return to the desktop, without interrupting the tracker, a message appears when the track is completed.

The downside of the automatic tracker is that there are no controls. If the track is being defeated by some aspect of the shot, you have no options, no adjustments.

We tested the automatic tracker against several leading industry stand alone 3D trackers and the results were surprisingly strong. While you have no controls in one case we managed in 5 minutes to get a better track than an experienced 3D tracker could achieve after 4 hours with boujou 2, and as mentioned above, in this test shot the autotracker completely outperformed the manual tracker. Unfortunately when it does fail – you are left wondering if every tracker would fail at this point – and there is no way to know this without attempting the shot in a range of other 3D trackers. You can manually edit the action setup, but not adjust the actual track solution.

Once the track was completed, we composited the green screen foreground over the now moving high resolution background by parenting the green screen foreground image under the new camera.