It not very beautiful as I haven’t found a good way to convert the matrices between the two worlds.
But it gives an idea on how is could work.

It would be really nice to see an example with a “real” matrix transformation, where focus and zoom of the 3d cameras fit together.
I have not seen examples of this anywhere yet.

I’ve seen a few examples of people transforming the whole world, but that is not as interesting.

Download the code here and give me a shout if you make progress!

The FLARToolkit is my own version, that you can see here.

Though if you want to track one marker it’s not the fastest.
The fastest tracker I’ve seen is the Alchemy branch of FLARToolkit.

I have combined the Alchemy version with some code from FLARManager.
I used the 3D engine converters and a snipplet for Smoothing the movements.
I have noticed that FLARToolkit is good at finding the position of a marker, but not always as good at finding the angles. Easy to figure ut why, but it becomes a problem if you want to place a non flat 3d object. It will flicker back n forth a lot.
To make this look better I have changed the smoothing algorithm to smooth the angle 4 times more then the position.

The youtube video looks more shaky than it does in reality. The framerate stays at a steady 30fps.

Please try it out HERE.
You will need to print this symbol first.

Download the sample and code HERE

To make the code work you will need MrDoob’s Stats and Away3D.

There are people making new Haar cascades which might be better trained for faces, so if you want try other cascades then the ones Mario and we are using, we have a simple PHP script that converts an XML file to the actionscript we are using. Just copy the results into /jp/maas/HaarCascade.as

You will need SimpleXML on your server for it to work. It usually comes with the php installation.

$xml = simplexml_load_file('haarcascades.xml');
$stages = $xml->haarcascade_profileface->stages->_;

for($s = 0; $s < count($stages); $s++){
    	$trees = $stages[$s]->trees->_;
    	if($s == 0)
    		echo "

tree = firstTree = new FeatureTree(". $stages[$s]->stage_threshold.");";
    	else
    		echo "

tree = tree.next = new FeatureTree(". $stages[$s]->stage_threshold.");";

    	for($t = 0; $t < count($trees); $t++){
    		$r1 = str_replace(" ", ",", trim(substr($trees[$t]->_->feature->rects->_[0], 0, strlen($trees[$t]->_->feature->rects->_[0])-1)));
    		$r2 = str_replace(" ", ",", trim(substr($trees[$t]->_->feature->rects->_[1], 0, strlen($trees[$t]->_->feature->rects->_[1])-1)));
    		$th = $trees[$t]->_->threshold;
    		$lv = $trees[$t]->_->left_val;
    		$rv = $trees[$t]->_->right_val;
    		if($t == 0)
    			echo "feature = tree.firstFeature = new Feature2Rects($th, $lv, $rv, [$r1], [$r2]);";
    		else if($t == (count($trees) -1))
    			echo "feature.next = new Feature2Rects($th, $lv, $rv, [$r1], [$r2]);";
    		else
    			echo "feature = feature.next = new Feature2Rects($th, $lv, $rv, [$r1], [$r2]);";
    	}
}

See it in action:

With better Blender 3D skills we would have made a Johnny Depp mask instead of a freaked out Gollum.

You can try it yourself HERE.

It’s an optimization of Quasimondos version of Ohtsuka Masakazu’s flash port of OpenCV.

What we have done to further optimize it is to use Flash10 specific functionality like using Vectors instead of Arrays. The original Othsuka/Quasimondo version was made for still images so we removed some calculations that became unnecessary while tracking a video. Another optimization was to look at casting between Int and Number (Int’s aren’t always faster if you need to cast them a lot for calculations).

- We will share the code here soon, we just have to prepare it for readability first.
- We will also make a more detailed description on how we did the optimization.

We used Away3D for the 3D rendering and wrote a really simple smoothing algorithm (the tracking results usually jumps around a lot).

For the mouth movement we simply track the sound level and open the mouth when it differs from the regular background noise.

In swedish only, read it here.

See a video of the result:

Try it yourself on:
www.t-post.se/rockpaperscissors

This was made in Flash (AS3.0 for Flash10) using FLAR Manager and Away3D.

Comment:
- Today is a momentous occasion. Today, for the first time in history,
a Japanese game show host looked at another country and said “WTF?”
At Gizmodo

- Det är otroligt fånigt. Men det är också så galopperande innovativt att jag inte vet var jag ska ta vägen.
At Expressen

Media:
Metro Magazine & pryl.feber.se

http://www.moment77.se/portfolio/flash/radiotjanst-tackfilm-gui/

We’ve disabled the save functionality.

Remember, this was built for internal use only, with strict guidelines. There is some bugs and it was written by speed, not perfection. So please keep this in mind when you play around. Enjoy!

Open Tackfilm GUI

Flash Developer Toolbox Essentials