Nachricht

Am Mittwoch, den 05.10.2011, hält Yixuan Shen um 16:30 im Raum P 1.4.17 einen Vortrag über seine Masterarbeit mit dem Titel:

Segmentierung von 3D-Oberflächendaten zur Analyse von Szeneninhalten in Innenraumumgebungen

Zusammenfassung:

In this work, the existing robot-framework of the GET Lab is to be extended by an analysis function i.e. detection, classification and reconstruction, of simple scene elements and objects based on geometric data. The geometric data to be considered are acquired by means of actuated tilted laser scanners (LRF) mounted on a real robot platform or by a 3D simulator, and are provided as 3D point clouds. These point clouds often contain more than 100,000 elements, which inhibits a direct interpretation of the scene. To examine the data of the point clouds with respect to shape and surface structure, a segmentation is to be applied. Within the segmentation process, any existing point is assigned a label in a way that those points belonging to the same region or surface are assigned to the same label. The segmented regions obtained this way shall then be used to approximate the raw data by a more abstract and adjustable representation. In this work, different point cloud features are introduced which serve as a basis for several segmentation techniques. A selection of segmentation methods is presented, whereas two suitable procedures are implemented in this work, namely normal- curvature segmentation and surface segmentation. For segmented point cloud regions an adequate representation is chosen considering both, the surface properties as well as the shape of the region. Finally, an analysis is performed to determine the neighborhoods and geometrical relationship of regions to one another. A series of experiments is conducted for the recognition of boxes in point clouds which embraces the segmentation and classification in simulated and real environments. The test results are visualized in a virtual reality where raw data, segmentation results and classified boxes are presented together.