I was asked by mail about markers and the tracking algorithm.
The automatic tracking in Kinovea works by computing the cross correlation coefficient between a candidate window and the feature window of the previous image.
For each possible position in the search window, we get a score, and the best score is the match (unless it is under a specific threshold in which case we assume the target was lost).
So it is a measure of how much the candidate look like the original.
- Contrasting area will make the matching easier. (marker should have a different brightness than its background. Bright on bright is not so good.)
- It is not invariant to rotation, so if rotating the marker changes its look, it will be harder to match. (triangle, rectangle, square: not so good)
- It will look in the immediate surroundings, so having a target that does not resemble any other part of its vicinity is better to avoid mismatches. (background and other parts on the person should be clear of anything that look like the marker).
To sum it up (based on the theory, I haven't done extensive testing in real conditions yet)
- Circular marker.
- Marker that takes about half the feature window.
- A color and brightness contrast with background that is not present in the rest of the search window.
Please post your findings, experiments, which markers work best, etc.
Anyone to do a video showing how various markers perform ?
