|
 |
Le 17-08-17 à 15:18, Bald Eagle a écrit :
> "Kenneth" <kdw### [at] gmail com> wrote:
>
>> 2) Perhaps the 'shoot-from point' (or from *multiple* points)
>
> I would think it would HAVE to be multiple points - because
> (a) if it were just a single point, then
> (1) it wouldn't make any sense, as there would only ever be one ray
> (2) and/or you'd get a solid wedge of sorts
> (b) if you only "scan" along that single vector - let's say, starting from any
> point along the other object's bounding box - then how do you handle the case(s)
> where the triangle being tested is perfectly parallel to the ray?
> How many intersection would it be calculated as?
The answer for this one is NAN (Not A Number)
>
> This suggests that a similar mechanism might be used in reverse to take a
> non-mesh object and construct a point cloud where the ray-[object]surface
> intersection points are. Then "just" triangulate the point cloud.
>
>
>
The test ray is shot from amy point that need to be tested to see wether
or not it's inside the mesh.
The case of a triangle been perfectly parallel to the vector do exist
and lead to artefacts. In that case, you need to slightly alter the
direction of the vector. Adding something like +0.000001 to one of the
components should be enough in most cases, but there are situations
where it can cause another triangle to become parallel to the vector.
Post a reply to this message
|
|
