A GPU Accelerated Signed Distance Voxel Modeling System

Abstract

This dissertation presents the capability of GPU-based parallel computing to support inter- active editing of solids represented by interpolated grids of signed distance values. The basic format of values on a grid is consistent with the format of an image stack that would be obtained from a volumetric imaging system such as magnetic resonance, positron emission tomography (PET), etc. or sent to a variety of 3D printing technology involving digital light projection (DLP) system or Powder bed and inkjet head 3D printing. Thus, this work represents another step toward the goal of a work ow from 3D scan to edit, analyse and fabricate all based on a uniform and robust data format. While interactive viewing of Signed Distance Field representation(SDF-rep) models has been previously demonstrated, here we present modeling operations that can be achieved in real or near-real time by taking advantage of GPU-based parallelism. Particular operations presented here include importing objects into the SDF-rep modeler, applying Boolean op- erations, sweeping cut as well as sweeping construction, manipulating the signed distance eld by offsetting SDF-rep models (i.e. uniformly manipulating signed distance eld to make the original SDF-rep models thinner or fatter without compromising the model integrity). Several applications integrated with those operations above give rise to the capability of modeling a shell version of solids with scaffold structure wrapped inside, which could be essential when printing bio-compatible scaffold material for implanting usage. In addition, this dissertation also presents a pipeline of computing discrete geometric skeletons from SDF-rep models, editing skeletal data, and re eshing (i.e. computing the distance grid, i.e. solving the eikonal equation, given partial information such as the skeletal data). I further illustrate(and present timings for) the skeletal editing procedures by provid- ing the example of bending a knuckle joint meanwhile adjusting the thickness of ngers on a hand model. Finally, we discuss the essentials of GPU-based parallel implementation and present the report about the efficiency of different options for memory management. Furthermore, in the body of work, the modeler system is capable for designing multi- material or continuous graded material models by applying user dened material-function on SDF-rep models for image-stack based 3D printing technology. Therefore, continuous gradation material property decoupled from geometry models is realized beyond the limi- tations of traditional boundary-rep based CAD software. Moreover, examples of creating a spatially controlled materials of various geometries are presented to help understand the whole pipeline. Last but not least, an approach for reconstructing 3D signed distance eld (SDF) model based on surface scanned point cloud data, is introduced here in order to enable the voxel modeler more versatile. Instead of only compatible with volumetric data or image stacks data, the 3D SDF reconstruction is desired for directly importing surface scanned point cloud and creating the signed distance eld version of the imported model. This 3D reconstruction method is also discussed in terms of efficiency when employing different amounts of cloud of points or grid sizes. A further discussion about the limitations of our 3D SDF reconstruction is presented to show the robustness of dealing with a few noisy outlier vertexes as well as the failure when importing the non-enclosed surface points data.