quote:Original post by rgirard413
Ok.. how bout, how do you incorporate collision detection?

...Poly poly1;p1.x = box->triangles[0].x;p1.y = box->triangles[0].y;p1.z = box->triangles[0].z;pOrigin = p1;... 

...Poly poly1;p1.x = box->triangles[0].x;<br>p1.y = box->triangles[0].y;<br>p1.z = box->triangles[0].z;<br>pOrigin = p1;<br>…<br> </pre> <br><br>???<br><br>My second observation is that your code is a bit hard to read. Why not use tabs? Instead of<br><br><pre><br>if {this and that)<br>{<br>do some stuff<br>do some more stuff<br>for (some loop)<br>{<br>inner loop stuff<br>}<br>}<br> </pre> <br><br>Do this:<br><br><pre><br>if {this and that)<br>{<br>  do some stuff<br>  do some more stuff<br>  for (some loop)<br>  {<br>    inner loop stuff<br>  }<br>}<br> </pre> <br><br>It will be easier to follow the logic that way. Its also a long function, and for debugging purposes (at least), you might consider breaking it into multiple functions with well-defined inputs and outputs. That way, it''ll be easier to find which piece is failing. The failure could be in the inputs or in calculating the outputs.<br><br>My third observation is that you''re using <b>very </b>  complex logic for the bounding box test. Did you build your logic based on one of the articles you''ve read? Its okay to use complex logic when it works, but the more complex the logic the less likely it is to work. Also be aware that it will be slower than it has to be. I suspect your code is going to be <b>very </b>  slow when you get it to work. One way to speed it up (a little bit) is to precalculate the triangle normals rather than calculate them every time you collide against the box. If its an aligned box, then you don''t even have to calculate them since they''ll be either (1,0,0), (0,1,0), or (0,0,1). If you use oriented boxes then you can still precalculate and just transform the normals when you transform the geometry.<br><br>Look at this part of your code:<br><br><pre><br>for(int i = 0; i < box->numtriangles; i++)<br>{<br>Point pOrigin;<br>Point p1,p2,p3;<br>…<br>// set some points<br>// calculate normal<br>// call a function called "intersect"<br>// maybe call "intersect" again<br>// maybe call "intersectSphere"<br>// maybe call "intersect" again<br>…<br> </pre> <br><br>Lots of "intersects*"''s in there. Here, it appears that you are walking through the triangles representing the box and checking your intersection point against each of the bounding box triangles. The <b>only </b>  reason I see to do this is if your bounding boxes are oriented bounding boxes (OBB''s). Even then this logic is complex. If you are using axis-aligned boxes then I would expect to see some very simple logic such as:<br><br><pre><br>if (sourcePoint.x > box.xmax || sourcePoint.x < box.xmin ||<br>    sourcePoint.x > box.xmax || sourcePoint.x < box.xmin ||<br>    sourcePoint.x > box.xmax || sourcePoint.x < box.xmin)<br>{<br>// no collision possible<br>}<br>else<br>{<br>// mesh did collide with bounding box. did deeper and collide<br>// point with full meshes.<br>}<br> </pre> <br><br>My written logic here is quick-and-dirty and a bit incomplete. I  mean, you''d want to test the entire line from sourcePoint to sourcePoint + velocityVector instead of just sourcePoint as I''ve shown. But the idea is to check against the global bounds of the bounding box, not against the individual triangles that represent a mesh of the bounding box. There isn''t really even any need to store a mesh of the bounding box. Just store the bounds and, if its an oriented box, the transformation matrix.<br><br>If you do have oriented bounding boxes, then you can still use the simpler logic. But you would have to transform sourcePoint and velocityVector into a coordinate system aligned with the oriented box before testing. But that transformation is simpler than the logic you''re using.<br><br>(There is a benefit to checking against bounding box triangles. You are at least already debugging the full triangle mesh collision test you will be doing if there is a collision with the box.)<br><br>I don''t immediately see why you are doing anything with ellipsoids and spheres. I suspect you are using a reference that I''m not familiar with.<br><br>Why not write out your logic—-not your code—-and we''ll see if there''s a flaw in your logic?  </i>   <br><br>Graham Rhodes<br>Senior Scientist<br>Applied Research Associates, Inc.