Which background substraction methods I should follow?

Question

I have a college work which I have a short video to cut the background off (keeping only what moves in the scene) with three differente method,

• Fixed Background or Fixed Fundus
• Average background or average fundus
• Median background or median fundus

So I searched and found this example from OpenCV: https://docs.opencv.org/3.4/d1/dc5/tutorial_background_subtraction.html So I guess this should be the first one

And this : https://learnopencv.com/simple-background-estimation-in-videos-using-opencv-c-python/ Should be the median fundus..

Am I right? I'm confused because the teacher did not pass us the terms in english. This algorithms are right with the methods required? Or where can I find example of these methods?

Answer 1

There are 2 types of background subtraction:

1. Morphological operation based
2. Artificial Intelligence(AI) based

Morphological operation based

This approach assumes that background is constant or better saying that it works fine if the background is constant. What I mean is that camera is stable.

• OpenCV library has already have some beautiful functions for this: BackgroundSubtractorMOG2 and BackgroundSubtractorKNN are some of famous ones. These functions are also able to hold story in memory so even camera is moving(background changement), they can update it.
• I also worked on this topic before, and designed an algorithm for that. Here is the github link for code and video results refered to that.

Note: In this approach, the biggest problem is lightening.

Artificial Intelligence(AI) based

This method is the popular one and used widely. Many meeting app in store are also using this one to change background. This approach basically detect the target objects and mask them. As an example you may have a look at the Mask-RCNN. So after detecting mask of target(human, car etc.), its very easy to change background. Camera movement and background changings don't affect.