Visual Scene Interpretation
Visual scene interpretation is an important ability for intelligent systems. We realize this ability as a complete loop from extracting bottom-up information from visual data, fusing these information with top-down knowledge, and improving bottom-up processing by feeding back interpretations as relevance or context information.
The goal for artificial vision systems is to analyze the visually perceivable environment in order to build up a rich scene representation, enabling the robot to navigate and manipulate its environment. In particular, we focus on determining salient spatial regions and develop local and global features for object and supporting structures detection , scene category learning , and functional property analysis. We work on representations which integrate top-down knowledge like existing models or information gathered in interaction with humans, e.g., spatial descriptions  or activity patterns .
We aim at analyzing these representations to extract semantic meaningful information to enhance the bottom-up processing. For example, identifying the scene category, the support type, or typical reference frames of a spatial structure improve detection of objects. Observed scene activities can be used to determine object affordances, knowing the layout of objects will provide the best candidate among a given set of grasping points.
Robotic grasping and processing of highly deformable objects, such as laundry, requires a very detailed object analysis. We support this task by developing algorithms for several steps of this process, including visual category recognition and grasp point detection for clothes.
- SFB 673: Alignment in Communication
- Spitzencluster: Intelligente technische Systeme OstWestfalenLippe
2014 | Journal Article | PUB-ID: 2563576A Detailed Analysis of a New 3D Spatial Feature Vector for Indoor Scene ClassificationPUB | PDF | DOI | WoS
Swadzba A, Wachsmuth S (2014)
Robotics and Autonomous Systems 62(5): 646-662.
2011 | Conference Paper | PUB-ID: 2034745Indoor Scene Classification using combined 3D and Gist FeaturesPUB | DOI
Swadzba A, Wachsmuth S (2011)
In: Computer Vision - ACCV 2010. Lecture Notes in Computer Science, 6493. Berlin, Heidelberg: Springer: 201-215.
2009 | Conference Paper | PUB-ID: 1890374A Computational Model for the Alignment of Hierarchical Scene Representations in Human-Robot InteractionPUB
Swadzba A, Vorwerg C, Wachsmuth S, Rickheit G (2009)
In: International Joint Conference on Artificial Intelligence. AAAI Press: 1857-1863.
Recent Best Paper/Poster Awards
Philippsen A, Reinhart F, Wrede B (2016)
International Conference on Development and Learning and on Epigenetic Robotics (ICDL-EpiRob)
Richter V, Carlmeyer B, Lier F, Meyer zu Borgsen S, Kummert F, Wachsmuth S, Wrede B (2016)
International Conference on Human-agent Interaction (HAI)
Carlmeyer B, Schlangen D, Wrede B (2016)
International Conference on Human Agent Interaction (HAI)