Knowledge-based modeling of laser-scanned trees

The increasing availability and power of range scanners has enabled us to scan larger and more complex objects. Trees, however, pose special problems for such scanning. Unlike buildings, trees have relatively more complex geometry, and the scanned tree data is inherently incomplete due to limited scanning resolution and occlusion from leaves and branches. Standard mesh generation techniques often fail because of this inadequate data. Our purpose is to use knowledge about the structure of trees and allometric theory to produce full polygonal meshes from the point clouds obtained from the scans. We aim to produce a mesh that plausibly recreates the tree that was originally scanned, but with extrapolated details that were missing in the point cloud. These surrogate models can then be used in place of the original point clouds for rendering.