Always play back your shot and plan which points you should track. This step is critical in order to provide track points useful to the solver. You can use the Playblast or Time Slider to play the shot. For more information about playback tools, see Live playback tools.
To solve for camera movement, the track points must be stationary. To solve for object movement, the track points must be fixed relative to the moving object. For example, if you are tracking the movement of a car, a feature on a headlight would help, but not a feature on an antenna waving in the wind.
When you solve, you can get improved results using survey constraints, so choose features with constraints in mind. The survey constraints are: Plane (such as a floor, table, or wall), Point (a 3D coordinate, such as a point corresponding to a position on reference geometry), Distance (measurement between points), and Depth (distance from a point to the camera). For more information, see Constraint strategies.
The more frames the feature is visible, the better. For example, in a 100 frame shot, it should be visible for at least ten to twenty frames. Also, avoid features that come in and out of view too much.
The solver needs a “balanced diet” of features from various areas, not just features grouped in one region.
In addition to tracking a variety of areas, you may want to track additional features around the intended 3D object location. If you intend to import the 3D model into the scene, look for points that clearly correspond to points on the model.
For example, if the shot is a slow pan from left to right, choose features in the center that overlap the start and end of the shot.
To establish a lock on the target, the tracker uses both the object’s shape and contrast with the background. Tracking artificial markers is helpful, but natural features work as well, such as tree branches, rocks, and windows. An object that has a definable tip can be easier to track.
In general, tracking larger objects gives more accurate results than tracking small objects. However, a larger target means the tracker may fail sooner due to pattern changes.