Brian Curless and Marc Levoy
|original||painted||1st scan||48 scans||final model||hardcopy|
We also have a 4-minute video describing how the buddha was scanned. Available formats are:
This page shows some recent results from our project to build a 3D fax machine. In particular, it shows a Volumetric Method for Building Complex Models from Range Images, developed by Brian Curless and Marc Levoy. A paper by this name appeared in Siggraph '96 and is currently on-line as a web document. This page also shows a hardcopy of the polygon mesh fabricated using stereolitheography. The software used in the creation of this model, as well as the range data and computer model itself are currently available for download:
As an application of this technology, we have embarked on a multi-year project to create a high-quality 3D archive of the sculptures of Michelangelo for scholarly, educational, and commercial purposes.
To look at results from our other projects, browse through our list of current research projects. Or you can return to our home page.
Figure 3 is a Gouraud-shaded rendering of one range image of the statuette. The range image was acquired by a Cyberware 3030 triangulation-based laser range scanner that has been modified to employ a spacetime analysis of the reflected laser light (Brian Curless and Marc Levoy, ``Better Optical Triangulation Through Spacetime Analysis'', Proc. ICCV '95). This figure illustrates the limited and fragmentary nature of the information available from a single range image.
Figure 4 is a Gouraud-shaded rendering of a polygon mesh created by integrating 48 aligned range images of the statuette. As each range image was acquired, it was converted to a volumetric representation and integrated into a volume representing the entire model. These volumes contained 400 x 900 x 400 voxels, each measuring 0.25mm on a side. Run-length encoding was used to reduce memory use. An isosurface was then extracted from the volume, producing the polygon mesh shown in the figure. Acquisition, integration, and isosurface extraction took approximately 3 hours on a 200 MHz R4400 with 256 Meg of RAM. The mesh contains 2.4 million polygons. Some holes are visible where the laser could not reach the surface of the original.
Figure 5 is a RenderMan rendering of the final model after application of an automatic hole-filling algorithm. This algorithm adds "plausible surfaces" at frontiers between "unseen" and "empty" regions of space, which can be determined from the volumetric representation. To improve knowledge of empty regions, a backdrop (not shown in these figures) was placed behind the model and 10 additional range images were acquired, bringing the total number to 58. The final model contains 2.6 million triangles and is watertight - i.e. it has no holes or self-intersecting surfaces. This property is necessary in order to fabricate the model (see figure 6). The rendering is of an 800,000-polygon decimated version of the final model, lit to simulate the appearance of the original before it was painted.
Figure 6 is a digitized video image of a hardcopy of the computer model - i.e. a 3D fax. It was manufactured using stereolithography (SLA) by 3D Systems Inc. To make it, we emailed the 800,000-polygon version of our model to their offices in Valencia, CA. They converted the model to a stack of about 500 filled planar contours spaced 150 microns apart. For each contour, they laid down a layer of liquid resin, then selectively hardened the layer with a scanning laser to reproduce the filled contour. Our buddha was fabricated lying on his back, beginning with the robes and ending with the stomach. This process took about 10 hours. Support for downward facing surfaces, in the form of additional webbing, is included in the contour stack, then filed off after fabrication. The fabricated model was then sanded and bead-blasted to smooth it. For didactic purposes, we asked them to leave the model unsmoothed in one small region. Here, the contours - nearly invisible to the naked eye - can be seen through a magnifying glass or in this close-up. For comparison, here is the corresponding portion of the decimated polygon mesh. For closeups of a single range image and an integrated mesh, look at some meshes from our model of Michelangelo's statue of David.