i
i
i
i
i
i
i
i
14.7. Level of Detail 691
ing some hysteresis around the r
i
value [661, 1079]. This is illustrated in
Figure 14.30 for a range-based LOD, but applies to any type. Here, the
upper row of LOD ranges are used only when r is increasing. When r is
decreasing, the bottom row of ranges is used.
Other Selection Methods
Range-based and projected area-based LOD selection are typically the most
common metrics used. However, many other are possible, and some will be
mentioned here. Besides projected area, Funkhouser and S´equin [371] also
suggest using the importance of an object (e.g., walls are more important
than a clock on the wall), motion, hysteresis (when switching LOD the
benefit is lowered), and focus. The viewer’s focus of attention can be an
important factor. For example, in a sports game, the figure controlling
the ball is where the user will be paying the most attention, so the other
characters can have relatively lower levels of detail [661].
Depending on the application, other strategies may be fruitful. Overall
visibility can be used, e.g., a nearby object seen through dense foliage can
be rendered with a lower LOD. More global metrics are possible, such as
limiting the overall number of highly detailed LODs used in order to stay
within a given polygon budget [661]. See the next section for more on this
topic. Other factors are visibility, colors, and textures. Perceptual metrics
can also be used to choose a LOD [1051].
14.7.3 Time-Critical LOD Rendering
It is often a desirable feature of a rendering system to have a constant
frame rate. In fact, this is what often is referred to as “hard real time” or
time-critical. Such a system is given a specific amount of time, say 30 ms,
and must complete its task (e.g., render the image) within that time. When
time is up, the system has to stop processing. A hard real-time rendering
algorithm can be achieved if the objects in a scene are represented by, for
example, LODs.
Funkhouser and S´equin [371] have presented a heuristic algorithm that
adapts the selection of the level of detail for all visible objects in a scene to
meet the requirement of constant frame rate. This algorithm is predictive
in the sense that it selects the LOD of the visible objects based on desired
frame rate and on which objects are visible. Such an algorithm contrasts
with a reactive algorithm, which bases its selection on the time it took to
render the previous frame.
An object is called O and is rendered at a level of detail called L,which
gives (O, L) for each LOD of an object. Two heuristics are then defined.
One heuristic estimates the cost of rendering an object at a certain level of
detail: Cost(O, L). Another estimates the benefit of an object rendered at