Adaptive Triangulation of Objects Reconstructed from Multiple Range Images
Abstract
When reconstructing objects from multiple range images one has to deal with the problem of large data sets. In our approach, the surface itself is implicitly defined by a signed distance function. The distance function is derived from the range images which are assumed to be already registered. A triangulated model is generated from a rough approximation of the object using the marching-cubes algorithm. Subsequently, the triangles representing the surface are adaptively subdivided until a pre-defined degree of accuracy is reached. Newly inserted vertices are back-projected to the original range images. Even in case of very large data sets, our approach pesents an efficient method of surface reconstruction due to the adaptive improvement of an initially rough mesh to the desired degree of accuracy.
