Tracking is a crucial part of effects work, but mastering the details can easily trip you up. Michael Thompson offers expert tips to help you achieve great results

For such an important part of the visual effects process, tracking is probably one of the least understood areas. Compared with most other fields of VFX work, it has little reference material available for it, although most artists will be aware of tracking and may even have dabbled in tracking the odd camera or two when needed.
On a basic level, tracking is often seen as quite an easy stepping stone into other areas within VFX, but it’s actually a technical and specialised field. It can take a long time to become a fully accomplished matchmove artist.
While this article won’t give you answers for every problem or instant fixes for different shots, it should give you a few ideas about minimising the chances of having problems and smoothing out the whole process from start to finish. You’ll find that nearly every shot you track will have different needs and different problems. One of the most important skills in tracking isn’t what software package you know, but rather having the ability to problem-solve and think of different ways to do things.

1. Manual tracks
One of the most common mistakes people make when tracking a shot is to put down bad 2D tracks. Your 2D tracks are the direct information from which the camera solve is calibrated from. The more accurate the data you put in, the more accurate the final camera solve will be. It may be time-consuming, but it’s well worth spending the extra time making sure each 2D track you put down accurately tracks the correct feature on the plate. Depending on which software you are using, play around with the likes of the search area, pattern type, accuracy thresholds and RGB levels to see how each affects the way the software tracks the features. There are no right or wrong settings: they will vary from shot to shot. One good technique is to try knocking the blue channel out of the image: this generally contains the most noise.

2. Wireframe display
When aligning your tracked camera to complex geometry in your 3D application, your viewport can become quite cluttered and difficult to work in. Backface-culling a wireframe object can help, but the viewport can still be confusing, particularly if you have multiple objects in the scene. Here’s good alternative to the backfaceculling option. This description is for Maya, but the same principles should apply for most 3D packages. Set your viewport to display geometry in shaded mode via Shading > Smooth Shade All. Next, enable the wireframe of the objects to also be displayed via Shading > Wireframe on Shaded. Select your Camera via View > Select Camera and then open the Attribute Editor with [Ctrl]+[A]. In the PerspShape tab under Environment, create your Image Plane. In the imagePlane tab’s Image Plane Attributes panel, set Alpha Gain to around 0.75. Under Placement, set Depth (of the Image Plane) to 1. You should now be able to see your geometry as a wireframe object through your Image Plane, but the object will occlude itself, making it easier to work with.

3. Determining point position in 3D space
When you’re creating geometry for your tracked camera, it’s useful to be able to use your calculated point cloud as a basis for the placement of the geometry. If you need to know the location of any point in the image sequence in 3D space, it’s surprisingly easy to work out. First, on one frame create a Null Locator. Looking through the camera, position it over the feature on the plate you want to know the 3D position of. Next, move the pivot point of the null to the same position as the camera. Move the playhead to a different frame so that the camera has moved position and the feature on the plate is still visible. Now scale the null until it sits back on the feature in the plate. This is the 3D location of the feature, which you can now use to correctly position your geometry.

4. Automatic tracking
Auto-tracking can be useful for generating tracks quickly, but its biggest problem is that the computer tracks everything in the plate – good and bad – and will also fail if motion blur is too strong. In the railway station image above, you can see a few typical tracks that would be generated by an auto-track. The green tracks represent examples of features that are good to use; the red tracks indicate the type of features that the auto track would pick up but you wouldn’t want to use. Any point down the sides of the lamp post that crosses the background would get picked up due to the high contrast in those areas – these are known as t-junctions, and are to be strictly avoided. The track at the base of the lamp post in the image has also been picked up. What appears to be a good track is really just the edge of a cylinder, and so will move around as the camera moves position. In this example, you might want to mask the lamp post out completely for an auto-track, then add a few manual tracks to features like paint chips afterwards.

5. Lens distortion
A lot of the time, you’ll find that tracking ‘distorted’ plates is perfectly fine – but on occasions, you’ll find that you might run into problems if you don’t account for lens distortion. The shorter the focal length of the lens is, the stronger lens distrotion becomes. It’s also more pronounced towards the edges and corners of an image. You can set your tracking software to try to work out the distortion of the image itself, but this can be inaccurate. Instead, use a lens grid, which enables you to accurately calculate how much distortion is being created. Create a grid of equal-sized squares. Print this as large as you can and mount it on a flat surface. Film this grid with the same camera and using the same focal length as in your tracking shot. Try to fill the whole frame of the image with the grid. Open the grid in your tracking software and manually set the distortion value until the grid lines run straight, note down the value, and use this value when tracking the real shot.

6. Scripting
Scripting can save you lots of time, especially when doing repetitive tasks. Here’s a simple MEL script to add cones to a point cloud. Select the points you want to use and run:

proc trackCone() {


The pp variable will carry the XYZ translation for each point:

float $pp[];


The pointCloud variable lists all the points you’ve selected:

string $pointCloud[]= `ls sl fl`;


Next a cone is generated for each point in the list:

for($eachObject in $pointCloud )
{

$pp = `pointPosition -w $eachObject`;
string $cone[]=`polyCone r 1 h 2 sx 4 sy 1 sz 0 n trackCone_1`;
move -ls 0 1 0 $cone.scalePivot $cone.rotatePivot ;
rotate 180 0 0;
move $pp[0] ($pp[1] – 1) $pp[2] $cone;
}

}
trackCone();

7. Manually editing the calibration
No matter how good your 2D tracks are, you’ll sometimes get jitter in the solved camera’s motion path. To improve camera motion and remove jumps, smooth the camera translations and re-compute the rotations in your tracking software. The rotations generally contain the high-frequency noise, particularly in the Z channel.

8. Body tracking
When doing a body track, a common mistake is to just key the characters rig every so often. To get a more natural result, add keys initially on all of the extreme poses, then go back and add where necessary.

9. Tracking for modelling
If there’s something you need to model, film or photograph the object and use the resulting point cloud to get proportions.

10. 3D for 2D
Export tracks from a solved camera to a 2D application for adding matte paintings and so on. Tracks from a solved camera can be more accurate and have less jitter.