next up previous contents
Next: 5.5 Other Methods Up: 5. Occlusion Culling Previous: 5.3 Building the Depth

5.4 Occlusion Testing

The algorithm for occlusion testing consists of two steps. First the screen space bounding rectangle of the potentially occluded object is computed and tested for overlap against the hierarchy of occlusion maps. If the occluders overlap the object, then a conservative depth value (minimum depth value) is computed for the screen space bounding rectangle of the candidate object. This depth value is tested against the the depth estimation buffer to determine whether the candidate is behind the occluders and is therefore occluded.

Each opacity value in a level in the occlusion map hierarchy corresponds to the coverage of the corresponding screen region. In the general case, the opacity values will range between 0.0 and 1.0 and values between the extrema correspond to screen regions which are partially covered by the occluders. To determine whether a candidate object is occluded, the overlap test is performed against a map level using the candidate's bounding rectangle. If the region corresponding to the candidate is completely opaque in the occlusion map, then the candidate is occluded if it lies behind the occluders (using the depth estimation buffer). The occlusion map hierarchy can be used to accelerate the testing process by starting at the low resolution maps and progressing to higher resolution maps when there is ambiguity.

Since the opacity values provide an estimation of coverage, they can also be used to do more aggressive occlusion culling by pruning candidate objects which are not completely occluded using a threshold opacity value. Since opacity values are generated using simple averaging, the threshold value can be correlated to a bound on the largest hole in the occluder set. However, the opacity value is a measure of the number of non-opaque pixels and provides no information on the distribution of those pixels. Aggressive culling is advantageous for scenes with a large number of occluders that do not completely cover the candidates, e.g. a wall of trees. However, if there is a large color discontinuity between the culled objects and the background, distracting popping artifacts may result as the view is changed and aggressively culled objects appear and reappear.


next up previous contents
Next: 5.5 Other Methods Up: 5. Occlusion Culling Previous: 5.3 Building the Depth
David Blythe
1999-08-06