There are a bunch of motion detection and argumented reality experiments with flash out there – most involve a webcam and are able to track a moving area quite accurate.

I was curious if flash would be able to handle the task of tracking multiple objects/(blobs.

Take a look at the result and track some cars in the Czech Republic (live).
.

The example is using Justin Windle´s “MotionTracker” approach, Andrei Thomaz´s port of the blobDetection-library from v3ga and Josh Chernoff´s version of the MJPEG streaming-component by Doug Hughe .

Basically two images/frames are filtered and compared. The deviating parts are filtered to show up as white “blobs”. This area is grabbed via getColorBoundsRect() and divided to be searched for individual blobs.

The Result has a slight amount of lag, resulting from the cams framerate, MJPEG-decoding of the webcam stream and the fact that it has to be fed via a proxy.

Also it´s much easier to get accurate results, if the background is known and fixed (no day/night changes or changing “static” objects like parked cars etc.).
And it would take bigger image source to get a more percise tracking but without further optimisations the limit lies somewhere at 600*x.

So in conclusion realtime multi-object motion detection with flash is possible!

Update:

Big thanks to Og2t for bringing Kanpei Baba´s genious approach to my attention. That guy just thought one step ahead as he harnessed the power of the player itself by using floodFill() to mark found blobs.

This is a perfect example how easy it is to oversee a great pattern – even while already using it!
Its just the same idea with getColorBoundsRect() - of course you could iterate through the pixels with a simple algo to get the same result – but why if the runtime does it – and so much faster than the AVM ?

The experiment has been updated and eats virtually ‘no’ cpu-time in comparison to the original.

Update2:

The cam I used until now has been moved to face away from the road – the feed of the new one is till a but too small and lacks quality but its hard to find a relyable realtime stream that captures moving objects, shows a suitable envirionment AND offers the right resolution.