Archived

This topic is now archived and is closed to further replies.

ZiM

Strange problem with Möller ray/triangle intersection test

Recommended Posts

Möller's ray/tri intersection test indicates an intersection when certain values are used, although in reality there shouldn't be any (drew it on paper to verify this); Ray origin is (48, 32, 64) Ray direction is (-48, 32, 0) Triangles vertices, counter-clockwise: (128, 64, 128), (64, 64, 108.8), (64, 64, 0) Debugging reveals that the determinant value from first dot product seems to be wrong. When I place breakpoint right after it debugger says det=222822. and calculating it by hand with calculator gives 301465.6 - could it be that this is too big value for floating point representation?
static bool RayTriIntersect(Vector &o, Vector &d, Vector &v1, Vector &v2, Vector &v3, Vector &ptHit)
{
	// Computes the intersection point of given ray and triangle if there is one.
	// Returns: true  if there is intersection and stores the isect point in ptHit
	//          false if given ray and triangle do not intersect
	Vector s, p, q, e1, e2;
	float  u, v, t, det, detInv;

	e1 = v2 - v1;
	e2 = v3 - v1;

	p = Vector::CrossProduct(d, e2);
	det = Vector::DotProduct(e1, p);

	if(det < M_EPSILON)
		return false;

	detInv = 1.0/det;

	s = o - v1;	
	u = detInv * Vector::DotProduct(s, p);

	if(u < 0 || u > det)
		return false;

	q = Vector::CrossProduct(s, e1);
	v = detInv * Vector::DotProduct(d, q);

	if(v < 0 || u+v > det)
		return false;

	t = detInv * Vector::DotProduct(e2, q);

	if(t > 1) // Make sure it's not longer than the original ray
		return false;

	ptHit = v1*(1 - u - v) + v2*u + v3*v;
	return true;
}
   
[edited by - ZiM on October 14, 2003 5:28:12 AM]

Share this post


Link to post
Share on other sites
I've got the same result as the computer. det = 222822.4f

so I guess YOU are wrong


e1 = (-64, 0, -19.2)
e2 = (-64, 0, -128)
d = (-48, 32, 0)


p.x = 32 * -128 - 0
p.y = 0 - (-48 * -128)
p.z = 0 - ( 32 * -64)
p = (-4096, -6144, 2048)


det = -4096 * -64 + 0 + -19.2 * 2048
det = 222822.4


[edited by - oliii on October 14, 2003 12:12:54 PM]

Share this post


Link to post
Share on other sites
Seems like you were right. I get the same result now.

I managed to fix the problem by replacing
if(u < 0 || u > det) 
with
if(u < 0 || u > 1.0) 
, same for the u+v.

I wonder why the original code at http://www.acm.org/jgt/papers/MollerTrumbore97/code.html have these...

Share this post


Link to post
Share on other sites
maybe because they assumed the ray direction was normalised. I'll check the algorithm later, as I use it myself, but I always use it with normalised ray. If you are right, good find. You may want to check your changes with a non-orthogonal triangle.

[edited by - oliii on October 14, 2003 12:58:36 PM]

Share this post


Link to post
Share on other sites
if you assume that the ray direction is normalized then
ASSERT it using assert() or any other method.
That will save you a lot of pain in general.

Share this post


Link to post
Share on other sites