Speaker: Oliver Mattausch
University of Zurich
Tuesday, March 4th, 14:00, in BC 329
New Methods for Indoor Scan Processing and Online Reconstruction
Recent improvements in portable 3D scanners enable the acquisition of 3D point clouds of entire office floors in a matter of hours. Despite the advances in acquisition technology, effective processing techniques are still needed to transform these datasets into a high-level representation that can be used in practical applications, such as automatic floor plan generation or interior design. The major challenge in indoor room scans is the presence of clutter and heavy occlusions, as well as the size of the acquired raw data of a typical indoor scan. I will talk about recently developed methods that allow the segmentation of such office floor scans into individual rooms and the extraction of the room architecture as watertight polyhedrons, followed by an automatic detection and classification of the interior objects. The availability of modern online acquisition devices has also created a demand to performing instant surface reconstruction of dynamic point data. A major challenge for a dynamic pointbased reconstruction is the robustness with respect to noise and outliers, as is commonly given in real-world scanning scenarios. Robust techniques can deal with outliers but are typically too expensive to achieve interactive reconstruction times for at least moderately sized point sets. I will discuss how this goal can be achieved using a novel continuous formulation of the robust weighted locally optimal projection (WLOP) operator based on a Gaussian mixture describing the input point density. Despite being significantly faster, our evaluations show that this method in general provides better quality than WLOP.
Oliver Mattausch is currently employed as a post-doctoral researcher in the VMML (Visualization and Multimedia) Lab of the University of Zurich, working in the DIVA (Data Intensive Visualization and Analysis) project on the reconstruction and visualization of large datasets.
He received his M.Sc. in 2004 from the Vienna University of Technology under the supervision of Eduard Meister Gröller on the topic of volume visualization on BCC lattices. He received his Ph.D. in 2010 from the Vienna University of Technology under the supervision of Prof. Werner Purgathofer and Michael Wimmer, with the main focus on visibility computations for real-time rendering. Before joining the VMML group in Zurich, he worked as a researcher at The University of Tokyo/ ERATO in the group of Prof. Takeo Igarashi with a JSPS (Japan Society of the Promotion of Science) Post-Doctoral Fellowship. He has published technical papers and has held courses at conferences like ACM Siggraph and Siggraph Asia, EG, EGSR, and ISMAR.
His current research interest is geometry processing for the purpose of reconstruction, visualization, and automatic understanding of real-world scanning data, in particular of largescale indoor scans.