> Mustapha Bismi:

> A "long time ago" (2000, not that far hopefully), I implemented it using the covariance matrix (Sorry, the work was explained in french only - http://mustapha.bismi.free.fr/articles/obb.pdf ).

Thanks, guess I’ll have to brush up on my French J

In any case, that looks like another vote for the convex hull being a potentially important step in the covariance approach.


> --andy

> Likely this is also brute-force, but John writes wicked-fast code, so it might be acceptable.


Hehe, this looks a lot like my previous code (except it’s a bit smarter than mine :)

This kind of approach has worked very nicely for us in the past as an offline process, but I need something faster now because it needs to be able to update while the app is running.


In fact, I’m willing to trade off quite a lot on the closeness of the fit (these OBBs will only be used for frustum-cull tests), so maybe a brute-force approach trying just a few coarse orientations would be good enough and faster than anything else. I guess I’ll just have to buckle down and implement the different approaches and compare them J