Smoothing Wavelet Reconstruction

This thesis present a new algorithm for creating high quality surfaces from large data sets of oriented points, sampled using a laser range scanner. This method works in two phases. In the first phase, using wavelet surface reconstruction method, we calculate a rough estimate of the surface in the form of Haar wavelet coefficients, stored in an Octree. In the second phase, we modify these coefficients to obtain a higher quality surface.

We cast this method as a gradient minimization problem in the wavelet domain. We show that the solution to the gradient minimization problem, in the wavelet domain, is a sparse linear system with dimensionality roughly proportional to the surface of the model in question. We introduce a fast inplace method, which uses various properties of Haar wavelets, to solve the linear system and demonstrate the results of the algorithm.