Outline of our program

This post is a simple step by step description of the main points in our program.

Setup scene
Build BVH over the objects
for #number of passes
  Trace the initial beams from the lightsources
  Calculate scattering events 
  Build BVH over the beams
  for #vertical pixels
    for #horizontal pixels
     Check for intersection with object BVH
      Collect intersected beams from the beam BVH
      calculate light from material
      calculate light contribution from beams

Scattering implementation

After implementing scattering in the participating media we had some problems when rendering highly scattering materials. Our first implementation used recursion, so with highly scattering materials we had a recursion depth of 100+. This caused a stack overflow in out program, which we solved by keeping a vector of beams we haven't checked for scattering yet, and then just looping trough that vector. The created scatterbeams are then added to the vector instead of recursively scattering.

void

Scene::addBeam(Beam beam) {

beamsToBeAdded.push_back(beam);

}

void

Scene::addBeams(){

int startSize = beamsToBeAdded.size();

//we run through the vector of beams that haven't been checked for scattering yet

while(beamsToBeAdded.size()>0){

Beam beam = (Beam)beamsToBeAdded[0];

beamsToBeAdded.erase(beamsToBeAdded.begin());

//we only scatter if the participating media is inhomogeneous or the      scattering value is greater than zero

//There is no reason to add the beams to the tree if the scattering coefficient is zero

if(hetero || beam.pm->getScatter(Vector3(0), beam.color) != 0){

if(beam.scatterDepth < maxScatterDepth) scatterBeam(beam);

//either subdivide the beam or add the current beam to the bvh.

if(useSubdivide){

subdivideBeam(beam);

}else{

beamvector.push_back(beam);

}

}

}

}

Another problem we encountered related to scattering was when combining scattering with the kd-tree for inhomogeneous participating media. When splitting the beams according to the kd-tree we had to make sure that only onescattering event occured across all the created split beams. This required some additional book keeping where each splitbeam knows the previous one, and can ask that beam if it's allowed to scatter.