Improving Completeness of Geometric Models from Terrestrial Laser Scanning Data

The application of terrestrial laser scanning for the documentation of cultural heritage assets is becoming increasingly common. While the point cloud by itself is sufficient for satisfying many documentation needs, it is often desirable to use this data for applications other than documentation. For these purposes a triangulated model is usually required. The generation of topologically correct triangulated models from terrestrial laser scans, however, still requires much interactive editing. This is especially true when reconstructing models from medium range panoramic scanners and many scan positions. Because of residual errors in the instrument calibration and the limited spatial resolution due to the laser footprint, the point clouds from different scan positions never match perfectly. Under these circumstances many of the software packages commonly used for generating triangulated models produce models which have topological errors such as surface intersecting triangles, holes or triangles which violate the manifold property. We present an algorithm which significantly reduces the number of topological errors in the models from such data. The algorithm is a modification of the Poisson surface reconstruction algorithm. Poisson surfaces are resilient to noise in the data and the algorithm always produces a closed manifold surface. Our modified algorithm partitions the data into tiles and can thus be easily parallelized. Furthermore, it avoids introducing topological errors in occluded areas, albeit at the cost of producing models which are no longer guaranteed to be closed. The algorithm is applied to scan data of sculptures of the UNESCO World Heritage Site Schönbrunn Palace and data of a petrified oyster reef in Stetten, Austria. The results of the method’s application are discussed and compared with those of alternative methods.

Abstract licence: CC BY

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.