Breakthrough / DVR-Scan

:vhs: Tool for extracting scenes with motion from videos (e.g. security camera or DVR footage). Written in Python, uses OpenCV.
http://www.dvr-scan.com/
Other
371 stars 49 forks source link

Investigate default kernel size for CNT algorithm #71

Open Breakthrough opened 2 years ago

Breakthrough commented 2 years ago

The latest version of DVR-Scan includes a new CNT algorithm, which has better performance at the expense of accuracy (and matches feedback from #48). It is probably worth doing some tests on the default kernel sizes to see if increasing them will improve accuracy:

Just to give some feedback here, CNT works fine, is indeed a little faster than MOG (~1.4× for SD to ~2.1× for Full HD), has more problems with brightness changes and lens flares, and for the closest results to MOG, seems to require a doubling of the kernel size (kernelSize = kernelSize * 2 + 1).

Originally posted by @ocram in https://github.com/Breakthrough/DVR-Scan/issues/48#issuecomment-991017809