Thanks for the nice work and making the source open.
Just spotted a (maybe minor) issue.
Seems that the issues originate from the original GMM source code.
In myGMM.cpp, line 458, var*var should be var. It's because var is already squared.
The update formula was derived for the variance (square of standard deviation).
Anyway, correcting it means tuning many parameters again... so I also don't bother with it...
In myGMM.cpp, line 455, the decision of background or foreground is based on
Hi, Kevin,
Thanks for the nice work and making the source open.
Just spotted a (maybe minor) issue. Seems that the issues originate from the original GMM source code.
In myGMM.cpp, line 458, var*var should be var. It's because var is already squared. The update formula was derived for the variance (square of standard deviation). Anyway, correcting it means tuning many parameters again... so I also don't bother with it...
In myGMM.cpp, line 455, the decision of background or foreground is based on
p(x | BG) > cthr
Hope this help.
AY