10 replies to this topic

[Lil_Lloyd]

I have a quadtree terrain, but nearby quads are being culled. Like so:

 

 

I was thinking maybe its my frustum extraction code? This is in OpenGL by the way in case matrix layout has any bearing on the problem and I am using the glm mat4x4 and vec4 types to represent my projection matrices and frustum planes 

 

void Frustum::Extract(const vec3& eye,const mat4x4& camMatrix)
{
	m_pos = eye;

	m_modView = camMatrix;
	m_projMatrix = GraphicsApp::GetInstance()->GetProjection();
	mat4x4 MVPMatrix = m_projMatrix * m_modView;

	for(int plane = 0; plane < 3; plane ++){
		int index = plane * 2;
		m_planes[index] = MVPMatrix[3] - MVPMatrix[plane];
		NormalizePlane(index);
	}

	for(int plane = 0; plane < 3; plane ++){
		int index = plane * 2 + 1;
		m_planes[index] = MVPMatrix[3] + MVPMatrix[plane];
		NormalizePlane(index);
	}
	
}

void  Frustum::NormalizePlane(int index){
  float normFactor = m_planes[index][0] * m_planes[index][0] + m_planes[index][1] * 
                     m_planes[index][1] + m_planes[index][2] * m_planes[index][2];
  m_planes[index] /= normFactor;
}

 

 

[Hodgman]

As a complete guess, try transposing your MVPMatrix matrix before extracting the planes from it.

I know when I did this, most of literature that I found was based around matrices that have been constructed to work with row-major vectors, like D3D traditionally did.

[Lil_Lloyd]

Thanks, I'll try that. This project is causing me a lot of trouble over simple things I understood the theory behind, but actually implementing it is so deceptively frustrating.

[Lil_Lloyd]

Nope that didn't work, but thanks anyway!

[Lil_Lloyd]

OK I just noticed something big, but fixing still hasn't solved the problem

 

void  Frustum::NormalizePlane(int index){
  float normFactor = m_planes[index][0] * m_planes[index][0] + m_planes[index][1] * 
                     m_planes[index][1] + m_planes[index][2] * m_planes[index][2];
  m_planes[index] /= normFactor;
}

 

 

Should have been

 

void  Frustum::NormalizePlane(int index){
  float normFactor = m_planes[index][0] * m_planes[index][0] + m_planes[index][1] * 
                     m_planes[index][1] + m_planes[index][2] * m_planes[index][2];
  normFactor = glm::sqrt(normFactor);
  m_planes[index] /= normFactor;
}

 

I was forgetting to squareroot the length! However I'm still getting weird errors...

[Lil_Lloyd]

ALSO - this may be a noobish question, but when I test a point against a frustum plane, do I have to normalise the point? Or leave it as it is? Additionally I have been using points defined in WORLD SPACE. This is correct yes?

[Polarist]

ALSO - this may be a noobish question, but when I test a point against a frustum plane, do I have to normalise the point? Or leave it as it is? Additionally I have been using points defined in WORLD SPACE. This is correct yes?

 

What do you mean by testing a point against a frustum plane?  Typically you test vectors against the sides of a frustum.  The concept of "normalizing a point" is kind of odd, typically you normalize a vector.  What space you use doesn't matter as long as you're testing everything in the same space.  So if your frustum is transformed into World Space, then that would be fine.  There's no "correct" vs "incorrect" here.

[Lil_Lloyd]

What do you mean by testing a point against a frustum plane?  Typically you test vectors against the sides of a frustum.

 

Well, vectors/points are things described by an x,y and z value so sometimes they can be easily confused at times unless you keep in mind except that vectors describe direction and magnitude and points, well, describe points in space! In 4d points and vectors the difference is the w component, points have a value of one and vectors have a value of zero, when you subtract a point from a point to get a vector the w component goes to zero as required.

 

As for testing points against planes as opposed to testing 'vectors' is that you do in fact test points such as corners of a cube etc. With that said, yes normalizing a point is indeed odd. 

[Lil_Lloyd]

RENDERING A FRUSTUM

 

So I found some code on rendering a frustum in old school style fixed pipeline GL

 

void Frustum::DrawFrustum(){
	mat4x4 temp = m_projMatrix * m_modView;
	mat4x4 inv = temp._inverse();
 
	vec4 fr[8]=
	{
		// near
		vec4(-1, -1, -1, 1), vec4( 1, -1, -1, 1), vec4( 1,  1, -1, 1),  vec4(-1,  1, -1, 1),
		// far
		vec4(-1, -1, 1, 1),	vec4( 1, -1, 1, 1),	vec4( 1,  1, 1, 1),  vec4(-1,  1, 1, 1)
	};
	// Transform all vertices:
	// multiply vertex array (fr) by matrix. result is transformed vertex array (tfr)
	
	for(int i = 0; i < 8; i++)
		fr[i] = inv * fr[i];
 

	for (int i=0; i<8; i++)
	{
		 fr[i].x /= fr[i].w;
		 fr[i].y /= fr[i].w;
		 fr[i].z /= fr[i].w;
		 fr[i].w = 1.0f;
	}

    glColor3f(1.0f,0.0f,0.0f);
	glBegin(GL_LINES);
	
		glVertex4fv(&(fr[0][0]));
		glVertex4fv(&(fr[1][0]));

		glVertex4fv(&(fr[1][0]));
		glVertex4fv(&(fr[2][0]));

		glVertex4fv(&(fr[2][0]));
		glVertex4fv(&(fr[3][0]));

		glVertex4fv(&(fr[3][0]));
		glVertex4fv(&(fr[0][0]));

		glVertex4fv(&(fr[4][0]));
		glVertex4fv(&(fr[5][0]));

		glVertex4fv(&(fr[5][0]));
		glVertex4fv(&(fr[6][0]));

		glVertex4fv(&(fr[6][0]));
		glVertex4fv(&(fr[7][0]));

		glVertex4fv(&(fr[7][0]));
		glVertex4fv(&(fr[4][0]));

		glVertex4fv(&(fr[0][0]));
		glVertex4fv(&(fr[4][0]));

		glVertex4fv(&(fr[1][0]));
		glVertex4fv(&(fr[5][0]));

		glVertex4fv(&(fr[2][0]));
		glVertex4fv(&(fr[6][0]));

		glVertex4fv(&(fr[3][0]));
		glVertex4fv(&(fr[7][0]));
	glEnd();
}

 

It seems solid to me, and I have a new 'debug' mode where I detach the camera from the frustum and can control it from above in an orthographic view. However, I'm getting behaviour that is odd. There is no frustum being rendered but only odd lines of varying length depending on camera rotation and sometimes the line doesn't even render unless the camera is in a certain location. Bizarre. 

 

Here's a video link/demo

 

Edited by Lil_Lloyd, 08 February 2013 - 09:26 AM.

[Polarist]

What do you mean by testing a point against a frustum plane?  Typically you test vectors against the sides of a frustum.

 

Well, vectors/points are things described by an x,y and z value so sometimes they can be easily confused at times unless you keep in mind except that vectors describe direction and magnitude and points, well, describe points in space! In 4d points and vectors the difference is the w component, points have a value of one and vectors have a value of zero, when you subtract a point from a point to get a vector the w component goes to zero as required.

 

As for testing points against planes as opposed to testing 'vectors' is that you do in fact test points such as corners of a cube etc. With that said, yes normalizing a point is indeed odd. 

 

Yes, I understand that.  The w is just there for convenience for certain transform matricies, that's also more of an implementation detail.

 

The distinction here is important, because there are situations where you would test a point against a frustum plane, and also times where you'd test a vector against a  frustum plane.  OR is it that you are trying to test whether a point lies within the frustum or not?  The question I responded to didn't make this clear.

[Lil_Lloyd]

As a complete guess, try transposing your MVPMatrix matrix before extracting the planes from it.

I know when I did this, most of literature that I found was based around matrices that have been constructed to work with row-major vectors, like D3D traditionally did.

 

You know I said this didn't work? Well, before I started normalising the planes properly it didn't. I tried it again last night and it worked PERFECTLY. However, it was only after agonising hours of experimentation that I decided to try this again. Including making a debug view from above with the bounding rectangles shown on screen. Taught me a few things though! Thanks.