Probabilistic Preference Measures for Knowledge-based Scene Interpretation (PRAESINT)

Project Data

DFG-Projekt Ne 278/9-1
2007 - 2012

Project team at the University of Hamburg:
Bernd Neumann (project co-leader)
Wilfried Bohlken

Project team at the Hamburg University of Technology:
Ralf Möller (project co-leader)

Project Goals

The project deals with the development of generic computer methods for the interpretation of everyday visual scenes, static or dynamic, such as indoor scenes for elderly assistance or traffic scenes for monitoring purposes. Scene interpretation of this kind requires both, extensive knowledge about everyday occurrences which can be represented by logic-based knowledge representation methods, and probabilistic models for uncertain interpretation decisions and predictions. In this project we investigate a particular way for integrating probabilistic models with formal knowledge representation for scene interpretation. Logic-based inferences delineate the space of consistent scene interpretations whereas probabilistic inferences determine preferred interpretations among those which are logically possible.