Interactive Augmented Reality in Realtime

[ThereRLG2006]

Btw, I make computer graphics, as you might see if you clicked on my name and looked at my page. Ever considered making a video game out of this? Picture this, an Augmented Reality Platformer! Instead of a joystick, you pick up a card and shove him around a virtually projected room. Even toss and catch him. Send me a message if you're interested, we'll talk more.

[mitebuster]

I remember a video available in youtube about that AR platformer. It can be improved of course. Anyway, what i mean is that tracking of movement of cards is not necessary. They are detectable directly since it is not an ordinary object (there's a pattern on it as i mentioned). User can use cards as a joystick. I studying for improvements such that you dont need a card to play. The non-stationary cam can move freely (but it should always see some reference points such as cards)

[mitebuster]

But virtual object can also get interaction with real objects (like pencil or hand.. or any moving object with respect to the any reference point) But, since the cam is moving, any moving object is not detectable easily (cos, background has a movement). My aim to distinguish the actual moving object regargless of changing, non-plain background. In the video, the background may seem "plain", but it is just a coincidence. The AR game idea is always splendid. But still we have some way to go.

[ThereRLG2006]

Of course, you wouldn't be able to track the movement of the card relitive to it's starting point if the cam moved around, at least not without some additional reference objects that don't move, like say another card laying next to it. But if the camera does not move, and the only movement is in the card, then you should be able to estimate how the card has moved from it's starting point, along with it's speed and angle. Should be easy to add physics at that point, right?

[ThereRLG2006]

Here's a thought. Have the computer assume that the starting position of the card is an infinite flat surface, and that the card is level. Then use Physics as soon as the card is picked up, so the object rolls off the card and appears to bounces on the surface where the card was resting. I don't think anybody's ever done that before.

[mitebuster]

The card has a pattern that is detectable by using statistical methods. The cam is calibrated for each frame and a fixed point on the card becomes the reference coordinate system. There are lots of known methods for it. When the card is moved, the coordinate system is transformed. Therefore, the virtual object is moved automatically. The main problem is to segment any moving real object on any moving background and calculate whether it collides a virtual object.

[mitebuster]

The idea is to estimate the location of real object with respect to the reference coordinate system. Detecting a movement through consecutive frames is easy. But if the background is changing or moving, detecting the movement of the moving real object is not so easy.