Level of Detail Skeletal Modelling

A desirable characteristic of any 3D model is knowledge of its physical properties. How does it bend, where is the weight located and so forth. One manner of addressing this problem is by the introduction of underlying skeletons. Each object is given and underlying "bone" structure describing solid fragments and their interconnection ("joints"). Using this information, physical simulations can determine what might happen if certain forces are applied to a particular part of the model, or how to move extremities if the center moves as well.

While this skeletal data would prove very useful in physical modeling and animation (consider describing animations only in terms of skeletal animaation, no explicit point movement), it is very complex. If a human were very close, it may be necessary to have articulated fingers or at least hands. However, when the person in question is on a distant mountaintop, having distinguished hands is a burden, as the data must be incorporated into the scene, but will not be visible. At greater distances we may wish to have different skeletons, where the skeleton is only accurate enough for the task at hand.

The first consideration might be to have a set of skeletons indexed by distance (or another metric). In most cases this is sufficient, as most models will have a spacial coherence property, meaning that for the most part, the skeleton will be at roughly the same distance from the viewpoint, ensuring a uniform skeleton. But what if we have a man on a mountain with a particularly long left arm (farfetched at first, but better examples to come). We couldn't care less about the fingers on his right hand, being several kilometers away, but his left hand could be mere centimeters from the camera. Here is where we introduce the LOD aspect. Consider the follwing model, where a model has a simple initial skeleton. Each bone in the skeleton has a test that can be performed to determine if a more accurate skeletal model is required. If so, the bone disappears and is replaced by the more accurate model. The result is a skeletal structure with high skeletal detail where it is required, and low detail where detail is not required.