Agnes Swadzba has received her Diploma in Computer Science from the Friedrich-Alexander-University of Erlangen/Nuremberg in 2006 and her PhD degree (Dr.-Ing.) from the Bielefeld University in 2011. She wrote her diploma thesis about "Estimation of Camera Motion from Depth Image Sequences" in cooperation with the company Elektrobit formerly 3Soft, Erlangen, and the Institute of Pattern Recognition (Lehrstuhl für Mustererkennung). Her PhD thesis "The Robot's Vista Space: A Computational 3D Scene Analysis" has been funded by the Collaborative Research Center 673 "Alignment in Communication" within the Project A4 "Alignment of Situation Models" and has been supported by the Applied Informatics group.

Currently, Agnes Swadzba is working as postdoctoral research follow in the Applied Informatics group. Her research interests are 3D computer vision, scene analysis, spatial awareness for robots, and robot learning through interaction. For more information see her Slides, posters, and movies presented at conferences and workshops.

Projects

The Holistic Scene Model:
Indoor Scene Classification using combined 3D and Gist Features
The holistic modeling realizes a categorization of room percepts based on the observed 3D spatial layout due to the fact the layout shows room type specific features and fMRI studies have shown that human brain areas being active in scene recognition are sensitive to the 3D geometry of a room.  In this work we show that indoor scene classification based on features capturing the spatial layout is possible. Class models have been trained on a new 3D-IKEA database containing dense 3D scans of 28 indoor rooms. read more ...
New: The 3D-IKEA database is now online available!!!

The Aligned Scene Model:
A Computational Model for the Alignment of Hierarchical Scene Representations
The ultimate goal of human-robot interaction is to enable the robot to seamlessly communicate with a human in a natural human-like fashion. Here, we propose a hierarchical spatial model that is empirically motivated from psycholinguistic studies. read more ...

The Articulated Scene Model:
Scene Analysis of Dynamic Scenes
We present a new system for a mobile robot to generate an articulated scene model by analyzing complex dynamic 3D scenes. The system extracts essential knowledge about the foreground, like moving persons, and the background, which consists of all visible static scene parts. In contrast to other 3D reconstruction approaches, we suggest to additionally distinguish between static parts, like walls, and movable objects like chairs or doors. read more ...

Teaching

Lectures

• WS 2011/2012: Lecture 3D Vision - From Perception to Semantic Scene Analysis (please visit the web page to get slides)
• SS 2011: Seminar Vision for 3D -- 3D for Vision (please visit the web page to get student talks)

Supervised thesis:

• Manuel Baum (Bachelor Thesis):
  Using spinImages for 3D object classification
• Patrick Artner (Diploma Thesis):
  Globally Consistent Registration of Time-of-Flight Camera Images
• Philipp Nitz (Diploma Thesis):
  Indoor Classification using 3D Data
• Sebastian Hammerl (Master Thesis):
  Detection of Functional Areas within Rooms using Depth Data
• Orhan Engin (Diploma Thesis):
  Integration of 3D Point Clouds in laser-based Navigation
• Christian Thöns (Bachelor Thesis):
  Implementation of an Intuitive Toolkit for Extrinsic Calibration of Swissranger Images and 2D Color Images