|
|
 |
 |
 | | | | |
 |
|||
 |
||||||
          |
 |
 |
||||
|
Accurate front-end culling using facet normals This technique performs a per-polygon test using the camera position and the polygon's facet normal and position to determine whether it is front or back facing. Its main advantages are its accuracy — the method culls the maximum number of polygons — and the fact that its calculations are done in object space, which eliminates vertices in the pretransform phase. The classic implementation of this technique requires us to precalculate and store the following information about the polygon: • The normal of the polygon (facet normal) stored in object space. (Some tools will export the normal, but otherwise it must be calculated using the polygon's vertices.) • A position on the plane of the polygon in object space — either one of the vertices or the polygon's center. • A list holding the indices of the polygon's vertices (indices list) To determine whether a polygon is front or back facing, we must first transform the viewer position (or camera position) to object space, calculate the viewing vector (the vector in which the viewer is "looking" at the polygon), and then calculate the dot product between the viewing vector and the polygon normal. If the result is positive, then the polygon is facing backwards. Here's an example of the traditional implementation for indexed triangles. The structure we're assuming for each face is 32 bytes in size, and is as follows: struct {
float x,y,z; } Vector; struct {
} faceData; //= total size 32 bytes Here's pseudo-code for the culling (please refer to the appendix for a C++ implementation): for ( each vertex in
pool )
mark vertex "not visible" } Transform Vw (viewer in world space) to Vo (viewer in object space) using the world-to-object transformation matrix. for ( each face in
array )
face->p1, face->p2 and face->p3 are indices to the vertex pool. Use them to mark the appropriate vertices "visible" } } |
|
|
Features
