Research Topic

Robust Clustering and Classification in Distributed Sensor Networks

Sensor networks, consisting of small low cost nodes with a sensing unit, a processing capability, a battery or secondary power supply, and a radio for wireless communication, are able to carryout complex signal processing tasks. In recent years, fully distributed sensor networks became a research topic of great interest with applications such as ambient intelligence, pervasiveness, monitoring and controling.

Recent research on signal processing for distributed sensor networks has introduced a new ICT paradigm which considers multiple hetrogeneous devices, such as smart phones, tablets, hearing aids or handheld cameras, that cooperate in multiple signal processing tasks (MDMT). This is radically different from current ICT paradigms, in which stand-alone devices focus on individual tasks or multiple devices perform a common signal processing task, as it is assumed in a classical wireless sensor network (WSN). Under this new paradigm, cooperation among the nodes can be beneficial when subsets of the nodes share common interests or observations. For cooperation to be successful, it is thus necessary to answer the question: who observes what?.

The networks that we consider in our research work are ad-hoc in the sense that there is neither a pre-defined network structure nor a centralized fusion center where information is gathered. In this scenario, the nodes communicate only within some neighborhood. A successful communication between the nodes is achieved by using decentralized labelling scheme that allows to uniquely identify every object of interest. Due to this decentralized mode of operation, an increased robustness againest single link failure is obtained, i.e., the network doesn't collapse entirely as would be the case for a fusion center failure in a centralized network. Furthermore, from a communication cost perspective, knowledge about which node sees which object allows the formation of interest specific clusters. Thus we consider in-network classification/labelling schemes where local information exchange only happens among single hope neighbors.

The research question we address is, to develop a distributed object labelling scheme in the context of camera networks. In a camera network setting where outliers can be commonly
encountered, the use of robust technique for the classification/labelling task is very important.

Key Research Area

Distributed Sensor Networks; Distributed Labelling

Contact

Freweyni K. Teklehaymanot
M.Sc.

Address:

Dolivostraße 15

D-64293 Darmstadt

Germany

Phone:

+49 6151 16 - 24384

Fax:

+49 6151 16 - 24404

Email:

teklehaymanot (at) gsc.tu...

 Print |  Impressum |  Sitemap |  Search |  Contact |  Privacy Policy
zum Seitenanfangzum Seitenanfang