This is a quick test I made regarding background-substraction in Flash (which is also called Scene building etc.). I wanted to test this because of another experiment which I have in mind. The result is very nice, and it is acceptable fast.
A simplified description of the used technique. First of all I have the WebCam and a Video. A (grayscale) Snapshot of the scene will be taken which is called background. The background should not include the things you want to extract. From that moment on you create a grayscale picture from the webcam and substract the background from it. I draw the video onto the screen and for every pixel that has a difference greater than 8 I copy a pixel from the original webcam image.
Have a look
[as]var frame: BitmapData = webcam.getCurrentFrame();
videoBitmap.draw( videoDisplay );
//– create grayscale image of webcam frame
buffer.applyFilter( frame, frame.rect, new Point, bt709 );
//– substract the background
buffer.draw( background, null, null, ‘difference’ );
//– apply threshold
buffer.paletteMap( buffer, buffer.rect, new Point, null, null, alphaMap );
//– create mask using alpha channel
frame.copyChannel( buffer, buffer.rect, new Point, BitmapDataChannel.BLUE, BitmapDataChannel.ALPHA );
//– draw
smallScreen.draw( videoBitmap );
smallScreen.draw( frame );
screen.draw( smallScreen, scale );[/as]
I updated the way the resulting picture is drawn (thanks André!). Therefore the alpha-channel is abused as a mask applied with some thresholding. Instead of BitmapData.threshold I use BitmapData.paletteMap with a pre-cached table because it is a little bit faster. It is much better using the alpha-channel as a mask because of smooth edges. One last problem: Using only the Y-channel of a picture results in to many holes in the resulting image. I will try the same with three dimensions instead of one.
You can avoid the copy pixels-loop with the BitmapData.threshold method. Or isn’t it faster ?
I will have a try on that. It would be some sort of masking since I have to copy every pixel from the original webcam image onto the screen if the difference is higher than a threshold value. It could work with an alpha channel. Cutting out the alpha using BitmapData.threshold and then one simple draw().
Hello guys, nice work…both of u :)
As i already introduced to joa i’m working on the project on the link, now i’m goin to move some features in flex, hope it will be nice…let m eknow what do u think about kilyka :)
More specific, i was thinking about use motion detection to extract a contour of subject and use it as bezier equation (…) to apply phisycs on it…what do u think about it ?
Would it be possible to get an example of this using loaded jpg data rather than the webcam?
I’m having difficulties making it work properly with just and image, and the example posted above is somewhat incomplete (values for alphaMap, etc)
Thanks.
Hi,
I would love to see your full source code for this, I have been working on a similar project, but having trouble. Did you extend the MovieClip class or the sprite class?
If you could send me your code, that would be amazing.
Thanks,
Naomi
Hi Joa,
I was wondering if you could offer a tip on the alpha map. I’m currently setting everything in the alphamap array halfway through (from index 127 upwards to 0xFFFFFF) with some decent results, but i think i could do better. Do you have any tips?
Thanks,
Dave
Yes, you should apply some morphological filters.
I’m working on a dynamic background subtraction algorithm that removes the static background from a webcam feed. I’m still using Flash 8 and was wondering if you could shed any light on my problem. I’ve captured a Bitmap of the static background and now wish to check each pixel of the video feed against that static bitmap. If the pixels are different, those pixels are displayed. Any tips on how to accomplish this without running through a huge loop for every pixel on the screen? I can provide code if needed.
Thanks for the great post!
I am workign in a project API called CamCanvas which is Flash Web cam to Canvas via JavaScript. And also trying to remove the background. The grayscale approach seems to be weak. I wanted to find a better approach, maybe also something that does the smooth edges?
http://www.taboca.com/p/camcanvas/background-remove/camcanvas.html
Hi ,
I need to remove background for hand motion detection ,.
after understanding your algorithm .
I coded up like this
I subtract consecutive gray frames from the 1st extract converted gray scale frame that considered as background or scene .
then results seems like this : http://imm.io/8nt8
Then i replace the difference meant motion value with the original image if the motion value is greater then 81 then replace original image with motion difference
and result like this : http://imm.io/8ntj
for the simple background subtraction , I followed this your blog link : http://blog.joa-ebert.com/2006/10/10/real-time-background-substraction/
My code after understanding your concept : http://codepad.org/qRa44GbM
Please suggest me if i understood your algorithm concept wrong.
I need to extract hand and remove background .
Thanks
Rabi