Claus SCHEIBLAUER

(Vienna University of Technology, Austria)

Purpose: In this talk we would like to present our system for editing and visualizing huge point clouds. It is a presentation of our current work-in-progress where we are already able to build, merge, edit, and visualize multiple point clouds with a size of up to 2 billion point samples per point cloud.
Methodology / Approach: Range laser scanning evolved as a means for documenting buildings or archeological excavation sites. Point clouds resulting from these laser scans can consist of hundreds of millions of points. To get a clean model from this vast amount of data needs several person months. Instead we try to visualize the data directly, so archeologists can have a quick overview of the already scanned areas (e.g., during a scanning campaign). The models in our viewer are combined point clouds from several scan positions, and it is possible to add new point clouds to an existing model. We also allow for deleting points from the model. Furthermore we developed a heuristic to estimate point sizes, which enables the viewer to display surfaces as closed objects without a special preprocessing step. For the point size heuristic it suffices to know the positions of the points.
Results: We built a model of the Domitilla catacomb which consists of more than 1 billion point samples and more than 1000 scan positions. The buildup process took some 6h 30min. The memory requirements of the model are the same as the memory requirements of the contributing point clouds, i.e., no memory overhead is introduced. In our system it is possible to execute the whole processing pipeline from reading in the point clouds, doing some editing, and building a model. Deleting small areas in the complete model, e.g. an unintentionally recorded scanner stand, lasts only some seconds. To delete an area of 60 million points takes about 4 minutes. We thank the FWF START project “Domitilla-Katakombe in Rom” and the ÖAW for access to the Domitilla scans.

Keywords: point-based rendering, out-of-core processing, range scanning, virtual reconstruction