This paper describes a fast algorithm, based on occlusion culling,
for rendering complex urban environments. Occlusion culling has been
widely used to achieve interactive frame rates for rendering very
large data sets. We extend previous work on occlusion culling by
taking full advantage of the visibility information, temporal
coherence, and graphics hardware. We describe a simple and efficient
algorithm, which exploits image-space approach and temporal coherence
to mark a large subset of invisible objects. Occlusion is calculated
dynamically and does not rely on any preprocessing of visibility
information or on a prior selection of occluders. As a result, it can
handle dynamic updates in the scene at a little cost. The approach
also provides a good framework for dealing with occlusion culling and
can be combined with many other current occlusion-culling approaches
to achieve better performance at a slight overhead in the computation
time.