Hello,

I need to make a very simple 3D collision detection, there won't be more than 10 objects in the screen but they can be very complex.

I generate a bounding rectangle/sphere based on the model but all the models can be imported, "glued" together and exported to make a new model (all of that using kinect).

My issue is that the models become quickly complex and don't fit much a rectangle or sphere but I need to make the bounding box or in this case "bounding boxes" on-the-fly, does anyone has a good algorithm or trick that could help me ?

Thanks,

rXp

What about using collections of bounding boxes/sphere (lists or trees) as your bounding shapes? If the number of glued objects is not too large, it can be quite feasible.

What about using collections of bounding boxes/sphere (lists or trees) as your bounding shapes? If the number of glued objects is not too large, it can be quite feasible.

Yes but what about complex .obj that are imported ? Like an animal...

What about using collections of bounding boxes/sphere (lists or trees) as your bounding shapes? If the number of glued objects is not too large, it can be quite feasible.

Yes but what about complex .obj that are imported ? Like an animal...

At that point, you may want to have your artists put in the bounding boxes in the model itself? or denote regions with tags or something.

That's my issue :( The .obj can be loaded by anyone and created or download from anywhere.

That's my issue The .obj can be loaded by anyone and created or download from anywhere.

Perhaps something akin to an Octree? Start with the largest bounding box possible, then subdivide it?

Hi.

Can you render the object to a texture at one colour, then convert your picking ray to a 2d texture point.??????

Hi.

Can you render the object to a texture at one colour, then convert your picking ray to a 2d texture point.??????

Not a bad idea, that would work well, if picking is their only collision.