An automatic method for eliminating spurious data from sensor networks

The operational benefits of network-centric data fusion systems are underpinned by the assumption of statistically consistent data fusion processes. This assumption may be severely tested when redundant and intermittently corrupted data is allowed to proliferate through the network. The challenge is thus to find a robust and unified solution framework. The paper presents such a framework, centred on the covariance intersection (CI) and covariance union (CU) data fusion algorithms. It reports a simulation-based evaluation of these algorithms, with respect to a grid network of sensors engaged in target tracking and track fusion. The network topology and the identity of corrupt data entries in the network are a priori unknown to the fusion processes. The performance of the combined CI/CU is measured with respect to its ability to eliminate the spurious data from the network automatically.