How does CppMT report confidence in its tracking after processFrame?
Opened this issue · 3 comments
What would be a good metric to use to return a confidence level 1 to 100% perhaps of how well the object was tracked?
The variable pointsActive can be used for this purpose by computing size(pointsActive) / size(pointsInitial).
I should soon find the time to implement this, but if you want to do it, this is a good line where to compute it: https://github.com/gnebehay/CppMT/blob/master/CMT.cpp#L182
Can I ask what you mean by "pointsInitial"? There doesn't seem to be any variable going by that name. Do you mean the set of keypoints used by the matcher?
What should happen to the confidence as the distribution of keypoints changes? I guess the algorithm assumes that the morphology of the object stays the same? What if the object gets occluded for some period of time?
Is this really a reliable approach ? What if there are high structure elements acting as distractors in the later frames that outnumber the size(points_initial). They will misguide the confidence calculation and cause boxes to be drawn on objects of non-interest.