Range Tracing

Abstract

In this report, we tackle the problem of merging an arbitrary number of range scans (depth images) into a single surface mesh. The mesh-based representation is superior to point-based approaches since it contains important connectivity information. Most previous mesh-based merge methods, however, lose surface details by using simplifying intermediate surface representations (e.g.\ implicit functions). Such details are essential for further processing steps, especially for feature-preserving reconstruction methods. Our method preserves all information (connectivity and the original measurement positions) as edges and vertices of a merged surface mesh. It avoids aliasing and smoothing artifacts, adapts to the local scanner sampling and is independent of the overlap size of the input range scans. The algorithm consists of only two basic operations and is therefore simple to implement. We evaluate the performance of our approach on highly detailed real-world scans acquired with different devices.