|
 |
While your method sounds very nice, I'm not sure how you are attacking some
of the practical problems... For example, for most primitive types you will
need to choose the set of planes to use for bounding. If you choose the
wrong set (eg. the 6 axis planes for a sphere) then your method really
isn't better that what I described. If you want a more accurate set of
planes you will need more (for a sphere you would need infinity), which
will make it slow. Also, you will need to define methods to create
efficient sets of bounding planes for each object if you want it to be
fast.
How are you doing the degeneration to a AB? It doesn't seem like it would
be easy...
Isn't a set of bounding planes called bounding slabs?
Andrew
Post a reply to this message
|
|
|
|
"Andrew Clinton" <ajc### [at] uwaterloo ca> wrote:
>While your method sounds very nice, I'm not sure how you are attacking some
>of the practical problems... For example, for most primitive types you will
>need to choose the set of planes to use for bounding. If you choose the
You are right, but this is the only object specific part of the
bounding planes. From then on (keeping track of transformations and
calculating the resulting AABB) everything is generic (independent of
specific object type).
>wrong set (eg. the 6 axis planes for a sphere) then your method really
>isn't better that what I described. If you want a more accurate set of
>planes you will need more (for a sphere you would need infinity), which
>will make it slow. Also, you will need to define methods to create
>efficient sets of bounding planes for each object if you want it to be
>fast.
My goal was to create and keep a 'good enough' AABB, not a perfect
one. Your sphere example: I just use the 6 axis planes. Sphere
intersection is so cheap that it doesn't matter if the AABB isn't
perfect.
>
>How are you doing the degeneration to a AB? It doesn't seem like it would
>be easy...
There is no longer a degeneration due to transformations.
Do you mean the calculation of an AABB from a set of bounding planes?
That is not so difficult, simply calculate the edge/corner of each
pair/triple of bounding planes.
>
>Isn't a set of bounding planes called bounding slabs?
May be, but as i used POV's 'plane' object i called it bounding plane.
Andreas
Post a reply to this message
|
|
