Tracking algorithms work on segmented objects which it tries to follow. ROI tracking has the detection phase build in itself. Binary tracking operates on already segmented binary objects (e.g. from threshold, spot detection, ...). The images intended for tracking should have well defined objects in order to be consistently segmented throughout their lifetime.
As in many cases this cannot be achieved during acquisition, some transformation (Contrast, Intensity Equalization, Background Removal) or preprocessing algorithms (such as Detect Regional Maxima / Minima or Detect Peaks / Valleys) should be used to improve object separation from the background. The document can be reverted back to original after being tracked with Revert to original Color image while maintaining the tracking result.
When the scene brightness is oscillating and ROI tracking is used it will automatically take into account changing LUTs settings to compensate for such oscillations. The keep auto scale feature should be turned ON.
When segmenting for binary tracking, special care should be taken of the number of objects, as this greatly influences the complexity of the problem to be solved. Many times it is better to have objects missing than too many objects. Missing object can introduce a gap into a track which can be later closed during tracking. On the other hand including “false” objects (dirt, debris of bigger, etc.) may cause bad tracking results because of high scene complexity.
Specifically, when thresholding, it is usually good to consider Cleaning, Smoothing and Filling Holes as well as setting restrictions on object size and possibly on object circularity as well.
When object segmentation is satisfactory the Timelapse must have a reasonable frequency (FPS - frames per second) for given object count and speed. The FPS may be lower when there is a low number of objects and therefore low ambiguity. However when the scene is dense the FPS requirement is much stronger. A good guess if any given time-lapse is well sampled is how well a human eye can assess individual object motion. If one cannot easily understand which object is which and where they are heading chances are that it is under sampled. Tracking results are always better with higher frame rate.