Determining Whether Caps are Necessary
As mentioned earlier, a completely closed shadow volume having a front cap and a back cap must be rendered whenever the camera lies inside the shadow volume or the faces of the silhouette extrusion could potentially be clipped by the near plane. We wish to render this more expensive shadow volume as infrequently as possible, so a test for determining when it is not necessary would be useful.
The near rectangle is the rectangle carved out of the near plane by the four side planes of the view frustum. As shown in Figure 6, we can devise a test to determine whether the shadow volume might be clipped by the near plane by constructing the set of planes that connect the boundary of the near rectangle to the light source. We call the volume of space bounded by these planes and by the near plane itself the near-clip volume. Only a point inside the near-clip volume can have an extrusion away from the light source that intersects the near rectangle. Thus, if an object is known to lie completely outside the near-clip volume, then we do not have to render a capped shadow volume.
Figure 6. The near-clip volume is bounded by the planes connecting the near rectangle to the light position L. If an object lies completely outside the near-clip volume, then it’s shadow volume cannot intersect the near rectangle, so it is safe to render it without caps.
When constructing the near-clip volume, we consider three cases: 1) the light source lies in front of the near plane, 2) the light source lies behind the near plane, and 3) the light source is very close to lying in the near plane. Let W be the transformation matrix that maps eye space to world space, and suppose that our light source lies at the 4D homogeneous point L in world space. We consider a point light source (for which LW=1) to be lying in the near plane if its distance to the near plane is at most some small positive value d. For an infinite directional light source (for which LW=0), we consider the distance to the near plane to be the length of the projection of the light’s normalized direction vector <Lx,Ly,Lz> onto the near plane’s normal direction. In either case, we can obtain a signed distance d from the light source to the near plane by calculating
If d>d, then the light source lies in front of the near plane; if d<-d, then the light source lies behind the near plane; otherwise, the light source lies in the near plane.
In the case that the light source lies in the near plane, the near-clip volume is defined by the planes K0=<0,0,-1,-n> and K1=<0,0,1,n>. These two planes are coincident, but have opposite normal directions. This encloses a degenerate near-clip volume, so testing whether an object is outside the volume amounts to determining whether the object intersects the near plane.
If the light source does not lie in the near plane, we need to calculate the vertices of the near rectangle. In eye space, the points R0, R1, R2, and R3 at the four corners of the near rectangle are given by
where n is the distance from the camera to the near plane, a is the aspect ratio of the viewport, equal to its height divided by its width, and e is the camera’s focal length, related to the horizontal field-of-view angle a by the equation e=1/tan (a/2). These four points are ordered counterclockwise from the camera’s perspective. For a light source lying in front of the near plane, the world-space normal directions Ni, where 0<i<3, are given by the cross products
where each R'i is the world-space vertex of the near rectangle given by R'i=WRi. For a light source lying behind the near plane, the normal directions are simply the negation of those given by Equation (19). The corresponding world-space planes Ki bounding the near-clip volume are given by
We close the near-clip volume by adding a fifth plane that is coincident with the near plane and has a normal pointing toward the light source. For a light source lying in front on the near plane, the fifth plane K4 is given by
and for a light source lying behind the near plane, the fifth plane is given by the negation of this vector. (Remember that if W is orthogonal, then (W-1)t =W.)
We determine whether a shadow-casting object lies completely outside the near-clip volume by testing the object’s bounding volume against each of the planes Ki. If the bounding volume lies completely on the negative side of any one plane, then the object’s shadow volume cannot intersect the near rectangle. In the case that an object is bounded by a sphere having center C and radius r, we do not need to render a capped shadow volume if Ki·C<-r for any i.
Figure 7 demonstrates that for point light sources, bounding volumes lying behind the light source from the camera’s perspective may often be mistaken for those belonging to objects that might cast shadows through the near rectangle. This happens when the bounding volume lies outside the near-clip volume, but does not fall completely on the negative side of any one plane. We can improve this situation substantially by adding an extra plane to the near-clip volume for point lights. As shown in Figure 7, the extra plane contains the light position L and has a normal direction that points toward the center of the near rectangle. The normal direction N5 is given by
and the corresponding plane K5 is given by
The plane K5 is added to the near-clip volume boundary for point light sources regardless of whether the light position is in front of, behind, or in the near plane.
Figure 7. Adding an extra plane to the near-clip volume for point light sources enables more objects to be classified as outside the near-clip volume.
See “For Further Information” at the end of this article for methods that can be used to determine whether other types of bounding volumes, such as ellipsoids, cylinders, and boxes, intersect the near-clip volume.