1 | initial version |
you might want to look at the author's paper
the shape / maximum of the psr changes with the features used:
the existing, heuristic threshold value is based on the min ground truth and the max false pos values of the test videos from opencv_extra (2 of them are grayscale)
Is there any recommended way of changing this parameter instead of just wild guessing based on empirical testing?
it's definitely not a percentual value, and it's not even linear. so i can't think of any other way, than improving your testing / heuristics, sorry ;(
there is a benchmark program here, if you happen to have ground truth data for your test video, you could use it for measuring.