Quick links

Registration and Matching of Large Geometric Datasets for Cultural Heritage Applications (thesis)

Report ID:
TR-820-08
Date:
April 2008
Pages:
127
Download Formats:
[PDF]

Abstract:

The last decade has seen an increasing number of projects to acquire detailed 3-D representations of cultural heritage objects at museums and archaeological excavations, with a goal of improving preservation, understanding, restoration, and dissemination. However, careful study and virtual reassembly of cultural heritage objects often requires sub-millimeter precision to faithfully capture fine details, irrespective of the size and number of objects. Existing 3-D scanning technologies can produce such detail for small models with a modest amount of manual labor but do not scale to the tens of thousands of fragments that may be present at an excavation. High-precision scanners also have limited viewing volumes, making it very difficult to acquire large objects such as statues.

Most scanning technologies used in cultural heritage acquire many raw 3-D scans, each from a single viewpoint. This data does not become readily usable until the relative viewpoints of each scan have been recovered, and the data is merged into a final model. Alignment, or registration, is the process of recovering these viewpoints, and is the focus of this thesis.
Assembling a large, fragmented object from its pieces involves recovering the pose of each fragment. We therefore examine the virtual reassembly problem as one of alignment.

We examine the alignment and assembly problems in cultural heritage scanning using data from the Digital Michelangelo and Theran Fresco projects. In the context of the Digital Michelangelo Project, which scanned many Michelangelo statues in Florence at approximately 0.25 mm precision, we address the challenges of aligning large, detailed range scans. Because of the statues' size, deformations due to calibration error are inevitable. We present an algorithm which accommodates warp in many large scans, thereby preserving the raw detail in the final model.

We also consider the case of many small range scans, in the context of the Theran Fresco project, which is using 3-D models of fresco fragments to aid in reconstruction. Although fragments contain few range scans, they lack the detail required for stable, automatic alignment using traditional techniques. We show how to exploit the properties of fresco fragments to obtain robust, automatic alignments, and to manually correct any misalignment in only a few seconds.

Finally, we present a method for matching fresco fragments based only on geometry. Many fragments contain no decoration or distinctive edge features, so exhaustively matching edge geometry between all pairs of fragments is essential. We show how this problem relates to range scan alignment, and present a new convolution-like algorithm for efficiently computing all possible alignment of each fragment pair.

Follow us: Facebook Twitter Linkedin