Decentralized data fusion and data harvesting framework for heterogeneous dynamic network systems

摘要

With the increasing number of available smart phones, sensor nodes, andnovel mobile smart devices such as Google glass, a large volume of datareflecting the environment is generated in the form of sensing data sources(such as GPS, received signal strength identification, accelerometer,microphone, images, videos and gyroscope, etc.). Some context-aware anddata centric applications require the online processing of the datacollected. The thesis researches on the decentralized data fusion and dataharvesting framework for heterogeneous dynamic network system consisting ofvarious devices with resource constraints. In order to achieve the flexibledesign, a general architecture is provided while the detailed data fusionand data exchange functions can be dynamically configured. A novel methodto use directed fusion graph to model the logical structure of thedistributed information fusion architecture is introduced. This directedfusion graph can accurately portray the interconnection among differentdata fusion components and the data exchange protocols, as well as thedetailed data streams. The directed fusion graph is then transformed into aformat with marked language, so that both human and machine can easilyunderstand and edit. In the field of data exchange protocols, this thesistargets energy-efficiency considering the resource constraints of thedevices and robustness, as the dynamic environment might cause failures tothe system. It proposes a refined gossip strategy to reduce retransmissionof redundant data. The thesis also suggests a design guideline to achievedifferent design aims for different applications. These results in thisfield can be integrated into the framework effortlessly. The configurationmechanism is another feature of this framework. Different from otherresearch work which consider configuration as a post-design work separatedfrom the main design of any middle-ware. This thesis considers theconfiguration part as another dimension of the framework. The wholestrategy in configuration sets up the foundation for the flexiblearchitecture, and makes it easy to adapt to the dynamic environment. Thecontributions in the above fields lead to a light-weight data fusion anddata harvesting framework which can be deployed easily above wirelessbased, heterogeneous, dynamic network systems, even in extreme conditions,to handle data-centric applications.