Collision detection for objects with unknown dimensions

[Soulus]

I'm working on a kind of markerless AR application, which allows user to load its own 3ds objects into the application. And after that he should be able to move new-added object around, with regards to the other objects on the screen. Thus i should somehow implement collision-detection for an object which dimensions are unknown to me. What is the best way to do it?

For now the only thoughts i have is to first get object's mesh, then get bounding box and then on this basis construct ellipsoid to use in collision-detection. Is this the only way, or i missed something and now doing overcomplicated things?

The best (and for now unreachable) outcome would be if i can place 2 objects one right close to another, without any buggy intersections in between.

[EgonOlsen]

For now the only thoughts i have is to first get object's mesh, then get bounding box and then on this basis construct ellipsoid to use in collision-detection. Is this the only way, or i missed something and now doing overcomplicated things?

The best (and for now unreachable) outcome would be if i can place 2 objects one right close to another, without any buggy intersections in between.

It doesn't sound overcomplicated to me...i'm not sure if this really fits your needs, because an ellipsoid can only be a approximation of the object, but the general idea seems fine to me.