Track objects with live video

This tutorial is a walkthrough to track points of interest in a live video.


To follow this tutorial, you need a camera. You can use your laptop webcam to perform a test, but it is not the best equipment to do so. Mainly because this camera constantly adapts luminosity.

When doing a tracking, the camera resolution is not the most important parameter. 600px wide is ok for most applications.

The latency of the camera is important. To compare latency, film a timecode in Millumin and take a screenshot to compare. Here for example, there is 14 frames of latency with a smartphone camera :


This is the most important part in tracking. The more difference of luminosity between the tracked points and the background, the easiest the tracking is.

Having a constant lighting will also make tracking more efficient.

The Effect

In your Millumin layer, go to the effect properties and add blob tracker effect :

In the settings, switch view-mode to threshold, and adjust the threshold parameter until only your point of interest persists :

Then, switch view-mode to blobs. The goal here is to be sure that your point of interest is never lost. You can adjust radius min and radius max for this purpose.

If you are tracking more than one point, you can go to the tracking view and adjust merge at parameter. Indeed, it will merge two blobs (consider them as one) if its centers are close. You can put this parameter to 0 to deactivate this behavior.
 Here is an example with a static image. At some point, increasing merge at tells the effect that those two points are only one :

You can now switch view-mode to tracking + original and test your tracking.

If ok, open the interaction-panel (CMD+M) and add some interactions on whatever you want (typically X and Y of a layer) :

To learn more about interactions, read this article : Interactions

Finally, you can downscale the analysis resolution by lowering the downsize parameter. This will lighten the process.

If the blob is a bit noisy, adjust the stabilization parameter. Keep in mind that the more you increase this parameter, the less the tracking is reactive.

You can download this project as an example :