Remove falsely detected foreground (image segmentation)

Hi,

I am working on moving object detection in video compressed domain. I have extracted compressed domain features from VVC reference software to create an image(frame).

However, the extracted image falsely detects some part of the background as foreground. For example, in a paragliding video, due to the movement of the camera the slightly changing background(trees) is also beaing detected as foreground. I want to eliminate that part and just have the segmentated image of the person in the parachute. As shown in 0001.png.

The orginal image, extracted image, binary image, morphologically closed image are as follows:

I used yanqi liu's code from: https://www.mathworks.com/matlabcentral/answers/1565216-how-to-remove-unwanted-portion-from-background as part of the process.

It seems to be the right way to go since it gets rid of the unwanted trees but for some reason the paragliding person is not fully detected.

I tried to change the value for the length in strel in imopen from 9 to 10, 12, 16, etc. but it gives me an error

bw = imopen(bw, strel('line', 9, 90));

Error using ~=

Matrix dimensions must agree.

Error in imseg_y (line 61)

bw(L~=ind) = 0;

I even tried to change the value for

bw = bwareaopen(bw, 100);

I changed it 500, 250, etc. I get the same error as above, no matter what value I choose.

Can anyone please help correctly detect the foreground while removing the falsely detected foreground?

Looking forward to everyone's suggestions! :)

P.S. I have attached my code for reference.

Answer 1

[Image Analyst]

Go here for better, and published, algorithms:

https://www.google.com/search?hl=En&as_sitesearch=www.visionbib.com%2F&bthG=Google%2BSearch&q=moving+object

[Yousra Ashfaq]

Thank you @Image Analyst. :) I have looked up research papers on google scholar and I have gone through their algorithms. I was wondering if you could give any suggestions based on my code and the problem I am facing here.

[Image Analyst]

I find it hard to believe out of all those papers you read, that this algorithm was the best. What paper was it from?

If you have the Computer Vision Toolbox, there are functions to detect foregrounds and moving objects:

https://www.mathworks.com/products/computer-vision.html