Semantic-level fusion of heterogenous sensor network and other sources based on Bayesian network

Information fusion systems that involve the use of heterogeneous sensor networks often face the problems of loss of data and uncertainty in data caused by vulnerability of networks where sensor nodes may be attacked or break down, limited bandwidth which may cause network congestion, and urban environments which may affect the sensor measurements. In this paper, we propose to address the above mentioned problem by employing information from other sources (e.g., textual situation reports, open-source web information, news reports and social media etc.) to augment estimation from physical sensors (e.g., video, acoustic, seismic, radar and multispectral data). A semantic-level information fusion (SELF) framework is developed based on Bayesian network, which is capable of (i) integrating information of different types (hard and soft data); (ii) incorporating contextual information and prior knowledge into the process; and (iii) dealing with loss of data and uncertainties inherent in all data sources. An adversarial event detection problem is used as an example to illustrate the effectiveness of the proposed system.