Erleben, Kenny and Somchaipeng, Kerawit and Sporring, Jon

**Abstract:**

A scale space approach is taken for building bounding volume

hierarchies for collision detection. An elliptical bounding volume

is generated at each node of the bounding volume hierarchy using

estimates of the mass distribution.

Traditional top-down methods approximates the surface of an object

in coarse to fine manner, by recursively increasing resolution by

some factor, e.g. 2. The method presented in this article analyzes

the mass distribution of a solid object using a well founded

scale-space based on the Diffusion Equation: the Gaussian

Scale-Space. In the Gaussian Scale-Space, the deep structure of

extremal mass points is naturally binary, and the linking process is

therefore very simple.

The main contribution of this paper is a novel approach for

constructing bounding volume hierarchies using multi-scale

singularity-trees for collision detection. The bounding volume

hierarchy building algorithm extends the field with a new method

based on volumetric shape rather than statistics of the surface

geometry or geometrical constructs such as medial surfaces.