Not sure how to "filter unstable movements" correctly #2
Description
In the pulse rate paper, they "filter out unstable movements". To quote the paper:
Many of the feature points can be unstable and have erratic
trajectories. To retain the most stable features we find the
maximum distance (rounded to the nearest pixel) traveled
by each point between consecutive frames and discard points
with a distance exceeding the mode of the distribution.
When we tried this, we found that most position deltas (when rounded to the nearest pixel)
were zero or one - especially if you do this step after the cubic spline interpolation step!
This meant that you ended up filtering out a large number of points!
At the time, we tried an alternative function which used the median, but we never used it.
At some point it might be worth trying this again and seeing this step is useful or not!