Webmaster Forum / Flash / General Flash Topics / July 2008

There are two things that pop to my mind which could be causing this.
Fortunately one solution will probably help with either problem.

The first is that webcams likely use some kind of compressing on the image.
Something like jpeg or the equivalent for motion. This would produce little
artifacts around the edges of things and that would show up in the moving part
of the image when you try and do your detection. Your camera may have settings
to control the quality of the image? So that could help.

The other problem that I would guess is that there is often signal noise in a
web came. These are just small off colored pixels that appear randomly all over
the place. Generally they are fairly low. The hard solution for this would be
to buy a more expensive web cam, and use liquid nitrogen or something to cool
the sensor. :)

Okay so here is the easy solution to both of the above. Use the convolution
filter. It sort of averages out little squares of pixels. So  any small
fragments will fade out and any large features will remain. Use it on your
final difference image, not the source images.

But then I have to wonder why you even need the motion detection. Unless you
are then going to pass along the detected motion for further handling, you
probably don't need to go to the trouble.

The flash Camera class has an activity level method where you can tell the
camera to be inactive until the level reaches a certain point. You can then
start your camera recording.

Who said anything about AS3? The AS2 camera class has a perfectly good
onActivity event handler.

And the convolution filter is available in AS2 as well. It really will help to
remove those artifacts. Here is my definition for the filter I used.

var rows:Number = 3;
var cols:Number = 3;
var matrixC:Array = [1, 1, 1, 1, 1, 1, 1, 1, 1];
var divisor:Number = 9*2;
var convolution:ConvolutionFilter = new ConvolutionFilter(rows, cols, matrixC,
divisor);

That was only defined once and then in the part of my code that looped was:

    oldData = newData.clone();
    newData.draw(myVideo,matrix);
    myBMP.draw(oldData);
    myBMP.draw(newData,new Matrix(),new ColorTransform(),"difference");
    myBMP.applyFilter(myBMP,new Rectangle(0, 0, 320, 240),new
Point(),convolution);
    myBMP.threshold(myBMP,new Rectangle(0, 0, 320, 240),new
Point(),">",0xff101010,0xff00FF00,0xffffffff,false);
    var myRect:Rectangle = myBMP.getColorBoundsRect(0x00ffffff, 0x0000ff00, true);

But again if you aren't planning on then taking and processing myRect further
(that is the rectangle that contains the different pixels) there really is no
point.