Particle Deposition Terrain Generation - c++

Question

I'm using Particle Deposition to try and create some volcano-like mountains procedurally but all I'm getting out of it is pyramid-like structures. Is anyone familiar with the algorithm that might be able to shed some light on what I might be doing wrong. I'm dropping each particle in the same place at the moment. If I don't they spread out in a very thin layer rather than any sort of mountain.
void TerrainClass::ParticalDeposition(int loops){
float height = 0.0;
//for(int k= 0; k <10; k++){
int dropX = mCurrentX = rand()%(m_terrainWidth-80) + 40;
int dropY = mCurrentZ = rand()%(m_terrainHeight-80) + 40;
int radius = 15;
float angle = 0;
int tempthing = 0;
loops = 360;
for(int i = 0; i < loops; i++){
mCurrentX = dropX + radius * cos(angle);
mCurrentZ = dropY + radius * sin(angle);
/*f(i%loops/5 == 0){
dropX -= radius * cos(angle);
dropY += radius * sin(angle);
angle+= 0.005;
mCurrentX = dropX;
mCurrentZ = dropY;
}*/
angle += 360/loops;
//dropX += rand()%5;
//dropY += rand()%5;
//for(int j = 0; j < loops; j++){
float newY = 0;
newY = (1 - (2.0f/loops)*i);
if(newY < 0.0f){
newY = 0.0f;
}
DepositParticle(newY);
//}
}
//}
}
void TerrainClass::DepositParticle(float heightIncrease){
bool posFound = false;
m_lowerList.clear();
while(posFound == false){
int offset = 10;
int jitter;
if(Stable(0.5f)){
m_heightMap[(m_terrainHeight*mCurrentZ)+mCurrentX].y += heightIncrease;
posFound = true;
}else{
if(!m_lowerList.empty()){
int element = rand()%m_lowerList.size();
int lowerIndex = m_lowerList.at(element);
MoveTo(lowerIndex);
}
}
}
}
bool TerrainClass::Stable(float deltaHeight){
int index[9];
float height[9];
index[0] = ((m_terrainHeight*mCurrentZ)+mCurrentX); //the current index
index[1] = ValidIndex((m_terrainHeight*mCurrentZ)+mCurrentX+1) ? (m_terrainHeight*mCurrentZ)+mCurrentX+1 : -1; // if the index to the right is valid index set index[] to index else set index[] to -1
index[2] = ValidIndex((m_terrainHeight*mCurrentZ)+mCurrentX-1) ? (m_terrainHeight*mCurrentZ)+mCurrentX-1 : -1; //to the left
index[3] = ValidIndex((m_terrainHeight*(mCurrentZ+1))+mCurrentX) ? (m_terrainHeight*(mCurrentZ+1))+mCurrentX : -1; // above
index[4] = ValidIndex((m_terrainHeight*(mCurrentZ-1))+mCurrentX) ? (m_terrainHeight*(mCurrentZ-1))+mCurrentX : -1; // bellow
index[5] = ValidIndex((m_terrainHeight*(mCurrentZ+1))+mCurrentX+1) ? (m_terrainHeight*(mCurrentZ+1))+mCurrentX+1: -1; // above to the right
index[6] = ValidIndex((m_terrainHeight*(mCurrentZ-1))+mCurrentX+1) ? (m_terrainHeight*(mCurrentZ-1))+mCurrentX+1: -1; // below to the right
index[7] = ValidIndex((m_terrainHeight*(mCurrentZ+1))+mCurrentX-1) ? (m_terrainHeight*(mCurrentZ+1))+mCurrentX-1: -1; // above to the left
index[8] = ValidIndex((m_terrainHeight*(mCurrentZ-1))+mCurrentX-1) ? (m_terrainHeight*(mCurrentZ-1))+mCurrentX-1: -1; // above to the right
for ( int i = 0; i < 9; i++){
height[i] = (index[i] != -1) ? m_heightMap[index[i]].y : -1;
}
m_lowerList.clear();
for(int i = 1; i < 9; i++){
if(height[i] != -1){
if(height[i] < height[0] - deltaHeight){
m_lowerList.push_back(index[i]);
}
}
}
return m_lowerList.empty();
}
bool TerrainClass::ValidIndex(int index){
return (index > 0 && index < m_terrainWidth*m_terrainHeight) ? true : false;
}
void TerrainClass::MoveTo(int index){
mCurrentX = index%m_terrainWidth;
mCurrentZ = index/m_terrainHeight;
}
Thats all the code thats used.

Answer 1

You should have a look at these two papers:
Fast Hydraulic Erosion Simulation and Visualization on GPU
Fast Hydraulic and Thermal Erosion on the GPU (read the first one first, the second one expands on it)
Don't get scared by the "on GPU", the algorithms work just fine on CPU (albeit slower). The algorithms don't do particle sedimentation per se (but you don't either ;) ) - they instead aggregate the particles into several layers of vector fields.
One important thing about this algorithm is that it erodes already existing heightmaps - for example generated with perlin noise. It fails miserably if the initial height field is completely flat (or even if it has insufficient height variation).
I had implemented this algorithm myself and had mostly success with it (still have more work to do, the algorithms are very hard to balance to give universally great results) - see the image below.
Note that perlin noise with the Thermal weathering component from the second paper may be well enough for you (and might save you a lot of trouble).
You can also find C++ CPU-based implementation of this algorithm in my project (specifically this file, mind the GPL license!) and its simplified description on pages 24-29 of my thesis.

Answer 2

Your particles will need to have some surface friction and/or stickiness (or similar) in their physics model if you want them to not spread out into a single-layer. This is performed in the collision detection and collision response parts of your code when updating your particle simulation.
A simple approach is to make the particles stick (attract each-other). Particles need to have a size too so that they don't simply converge to perfectly overlapping. If you want to make them attract each other, then you need to test the distance between particles.
You might benefit from looking through some of the DirectX SDK examples that use particles, and in particular (pun arf!) there is a great demo (by Simon Green?) in the NVidia GPU Computing SDK that implements sticky particles in CUDA. It includes a ReadMe document describing what they've done. You can see how the particles interact and ignore all the CUDA/GPU stuff if you aren't going for massive particle counts.
Also note that as soon as you use inter-particle forces, then you are checking approximately 0.5*n^2 combinations (pairs) of particles...so you may need to use a simple spatial partitioning scheme or similar to limit forces to nearby groups of particles only.
Good luck!