Streaming Surface Reconstruction Using Wavlets

We present a streaming method for reconstructing surfaces from large data sets generated by a laser range scanner using wavelets. Wavelets provide a localized, multiresolution representation of functions and this makes them ideal candidates for streaming surface reconstruction algorithms. We show how wavelets can be used to reconstruct the indicator function of a shape from a cloud of points with associated normals. Our method proceeds in several steps. We first compute a low-resolution approximation of the indicator function using an octree followed by a second pass that incrementally adds fine resolution details. The indicator function is then blurred using a modified octree convolution step and contoured to produce the final surface. Due to the local, multiresolution nature of wavelets, our approach results in an algorithm 10-100 times faster than previous methods and can process extremely large data sets in the order of several hundred million points in only an hour. CR Categories: I.3.5 [Computer Graphics]: Computational Geometry and Object Modeling—Curve, surface, solid, and object representations; Geometric algorithms, languages, and systems