SFB 627: Nexus
home uni uni suche search sitemap sitemap contact contact
unilogo Universität Stuttgart
SFB 627: Nexus
Project E3:

Distributed situation recognition based on evaluated context information

Deutsche Version
 
Prof. Dr. rer. nat. habil. Paul Levi

Universität Stuttgart
Institute of Parallel and Distributed Systems
Universitätsstraße 38
70569 Stuttgart

Tel. +49 (0)711 7816-387
Fax +49 (0)711 7816-250
Paul.Levi@ipvs.uni-stuttgart.de


Prof. Dr. rer. nat. Dr. h.c. Kurt Rothermel

Universität Stuttgart
Institute of Parallel and Distributed Systems
Universitätsstraße 38
70569 Stuttgart

Tel. +49 (0)711 7816-434
Fax +49 (0)711 7816-424
Kurt.Rothermel@ipvs.uni-stuttgart.de


Overview


Context information can be divided in two classes: observable context can be direclty captured by sensors in the real world; higher-level context (also referred to as a situation) may only be inferred by using external knowledge of direct context information. As an example you could conclude from the observable state of a room (e.g. number of persons present in the room, current noise level, state of the beamer etc.) if there is a meeting in the room. The identification of a situation heavily depends on individual interpretation and is always flawed by a certain inaccuracy.

The informational requirements of context-based applications are prevalently very specific and exceed directly observable context informations by far. For this reason this project develops general concepts in order to determine application-specific situations (higher-level context) with the available distributed environment model. This will enable future context-based applications to query arbitrary self-relevant situations. A suitable method for the recognition of situations will be automatically chosen and placed on the distributed nexus components based on the specification. As soon as a specific situation has been recognized the application will be notified of this.

In order to derive higher-level context this project intends to re-use already existing approaches of the AI field (e.g. Bayes networks, neuronal networks or situation graphs). In doing so the situation recognition has to be based on a substructure of distributed context information which is only available in varying degrees of quality. A great part of the context data is furthermore highly dynamic, e.g. the position information of mobile objects. Resulting from this is a set of essential questions regarding the efficient recognition of situations in a distributed system:
  • For efficiency and quality reasons it is necessary to perform the recognition as close to the distributed data sources as possible. Hence it will initially be analyzed how existing identification methods can be modularized and distributed on miscellaneous system components.
  • A further focus is on the quality of the extracted information. Because context information is only available in a distributed manner, changes dynamically and its quality (e.g. consistency, actuality, completeness) may vary strongly it is mandatory to apply a quantitive evaluation of the obversation quality for each situation. Suitable quality metrics for each applied situation recognition procedure need to be developed in co-operation with the Q project.
  • Furthermore the attributes of the distributed system strongly affects the observation quality (e.g. communication latency or degree of time sync difference). The magnitude of this can be significantly influenced by a system-wide distribution of the observation functions. In a further step the project will explore methods to distribute the situation recognition modules in the system to minimize processing cost (e.g. system load or message throughput). In doing so it is assumed that applications state the preferred quality of the observation additionally to the situation itself so that a minimum cost identification can occur.

Institute


- Institute of Parallel and Distributed Systems    
  Image Understanding Research Group    
  Distributed Systems Research Group    


People


- Project Heads    
 - Prof. Dr. rer. nat. habil. Paul Levi   
 - Prof. Dr. rer. nat. Dr. h.c. Kurt Rothermel   
- Research Assistants    
 - Dipl.-Inf. Oliver Zweigle   
 - N.N.