I am using this code for detecting abandoned objects in a video sequence but in image subtraction module the code detects the object initially and then stops detecting it once it stays in for a couple of frames?

foregroundDetector = vision.ForegroundDetector('NumGaussians', 3,'NumTrainingFrames',400);
videoReader = vision.VideoFileReader('C:\Users\Mahim Alam Khan\Documents\MATLAB\GMM #1\AOB.mp4');
se = strel('octagon', 6);
for i = 1:400;
    frame = step(videoReader);
    foreground = step(foregroundDetector, frame);
    end
filteredForeground = imclose(foreground, se);
filteredForeground = imopen(foreground, se);
blobAnalysis = vision.BlobAnalysis('BoundingBoxOutputPort', true,'AreaOutputPort', false, 'CentroidOutputPort', false,'MinimumBlobArea', 10);
% bbox = step(blobAnalysis, filteredForeground);
% result = insertShape(frame, 'Rectangle', bbox, 'Color', 'green');

videoPlayer=vision.DeployableVideoPlayer(10);
videoPlayer.Size='Custom';
videoPlayer.CustomSize=[650 400];
videoPlayer=vision.VideoPlayer('Position', [740, 200, 700, 400]);

videoPlayer1=vision.DeployableVideoPlayer(10);
videoPlayer1.Size='Custom';
videoPlayer1.CustomSize=[650 400];
videoPlayer1 = vision.VideoPlayer('Position', [20, 200, 700, 400]);
se = strel('octagon', 6);
while ~isDone(videoReader)
     frame = step(videoReader); 
     foreground = step(foregroundDetector, frame);
     filteredForeground = imopen(foreground, se);
     bbox = step(blobAnalysis, filteredForeground);
     result = insertShape(frame, 'Rectangle', bbox, 'Color', 'green');
     step(videoPlayer, result); 
     step(videoPlayer1, filteredForeground);
end

release(videoReader);

Answer 1

[Sammit Jain]

I think this is happening possibly because the image subtraction method you're using constantly updates the background too, which is why a moving object is detected, and not stationery ones (because a stationery one becomes a part of the scene in a later frame).

See if you can fix the background frame (that is, not update it) and it should work just fine.

All the best!

[Mahim Alam Khan]

Ok.I got the problem but how will I stop the background getting updated. Here I am using GMM for background subtraction and is there any way I can stop the background getting updated everytime making stationary objects to become part of background.

Answer 2

[Image Analyst]

You have to define how long a stationery part of the image can be in there before it's considered permanent. After all, if it's there long enough, how are we to know it's not just part of the furniture?

[Mahim Alam Khan]

Ok.I got the problem but how will I stop the background getting updated. Here I am using GMM for background subtraction and is there any way I can stop the background getting updated everytime making stationary objects to become part of background.

[Image Analyst]

The algorithms are presented here: http://www.visionbib.com/bibliography/motion-f743lu1.html Pick one and code it up.

[Mahim Alam Khan]

Ok thanks. But is there any way to tweak up the mentioned code in the question via some few lines of code. Please help..!!

[Image Analyst]

Try adjusting NumTrainingFrames. And you might have to generate a new foregroundDetector periodically at the longer elapsed time period that you want, like every 5 minutes or so.