Distributed Kalman Filtering using Broadcast Gossip

For many processes it is required to have a reliable view of an environment of interest. One way to achieve this is by performing a distributed Kalman filter. In this thesis, three distribution methods from different research backgrounds are implemented and evaluated using multiple metrics for use in a Gossip based wireless sensor network: consensus, weighted consensus and covariance intersection. The modular solution makes it possible to easily switch between the different implemented distribution methods and Gossip algorithms. From the evaluation based on metrics like the correctness of the estimate and the agreement among the different nodes, it follows that the naive consensus algorithm does not perform well. The weighted consensus and covariance intersection perform both with errors smaller than two degrees Celsius. However, the weighted consensus does require assumptions that covariance intersection does not.