Farias, Mitchell, and Silva [24,25] propose the ZSWEEP algorithm. Although it has its roots in sweep plane algorithms, ZSWEEP actually projects cells onto the viewing plane. The basic idea behind ZSWEEP is simple. First, ZSWEEP sorts the vertices of the mesh based on the current viewpoint. The algorithm then visits the vertices one at a time in back to front order. When it visits a vertex, ZSWEEP projects all faces attached to that vertex that are not yet projected. Using information stored in frame buffers, ZSWEEP can extract the parameters necessary for volume integration.
Of course, sorting vertex depths alone is not sufficient to create a proper depth ordering of the cells. However, tests performed by Farias, Mitchell, and Silva show that about 70% of the fragments projected are in front of all those previously projected, about 82% are behind no more than one other fragment, and over 99% are behind no more than two. Farias, Mitchell, and Silva conclude that, given their order of projection, they need in practice an A-buffer of only limited depth. Although A-buffers of even limited depth are not available yet on commodity graphics hardware, one could be implemented with high efficiency. Furthermore, ZSWEEP stands alone in accuracy and speed for software implementations of unstructured grid volume rendering.