Contributions to Conferences and Workshops of Agnes Swadzba

Assian Conference on Computer Vision (ACCV'10)
The presented paper proposes 3D features that allow successfully an indoor scene classification (e.g., kitchen, living room, ...) relying on the spatial layout of rooms. Capturing the spatial layout results in a scene representation that is independent from a detection of specific objects [paper, poster, database, projectpage].
 

International Conference on Robotics and Automation (IRCA'10)
The paper proposes an articulated scene model which is acquired by a robot during observation of a dynamic scene. As the scene is observed in 3D from one view point an important physic assumption is that the farest measurements along rays determine the static background while relocated objects will pop-out in front of the background as articulated scene parts [paper, talk, video, projectpage].

International Joint Conference on Artificial Intelligence (IJCAI'09)
The paper proposes a computational model for acquiring a hierarchical scene representation from verbal spatial descriptions and 3D perception of a robot's indoor Vista space. It follows the assumption that a hearer can build a model of a scene that is similar to the scene model the speaker has built from visual perception [paper, talk (IJCAI'09), poster (ICVSS'10), video, projectpage].

International Conference on Pattern Recognition (ICPR'08)
The paper proposes a method for detection and merging of redundant data in registered 3D point clouds. It is based on the alignment and projection of the data to/on a main plane which is used as a virtual image plane [paper, poster].

International Workshop on Statistical Pattern Recognition (SPR'08)
I presented an approach to recognize from a single 3D Time-of-Flight image the corresponding room type that can be seen on the image. The algorithm is based on  features computed from extracted planar patches in the given 3D point cloud [paper, poster]

CVPR'08 Workshop: Time-of-Flight Based Computer Vision (TOF-CV)
I presented work done in collaboration with Joachim Schmidt and Niklas Beuter about
"Tracking Objects in 6D for Static Scene Reconstruction" [paper, slides].

Dagstuhl-Seminar: Logic and Probabilistic for Scene Interpretation
Sven Wachsmuth and I presented a talk about "Probabilistic modeling of spatial and temporal aspects of scenes for Human Robot Interaction" [paper, slides].

International Conference on Computer Vision Systems (ICVS'07)
At this conference I presented the work done during my Diploma Thesis at the Chair for Pattern Recognition at the University of Erlangen-Nuremberg in cooperation with Electrobit (former 3Soft). A sequence of 3D point clouds from a Time-of-Flight sensor (from PMDTec) is aquired during an arbitrary camera flight and registered to a global point cloud [paper, slides].