Distributed terrain estimation using a mixture-model based algorithm

This paper describes a planar grid-based distributed terrain height estimation algorithm for use in a distributed data fusion sensor network. Each sensor node in the network represents the terrain using a Gaussian mixture to represent the elevation density in each grid cell. The local sensor node uses a rigorous probabilistic analysis of sensor measurement errors to associate individual measurements with multiple grid cells to account for in-plane uncertainty. Representing the elevation density in each grid cell as a sum of Gaussian distributions leads to a convenient channel filter implementation for distributed data fusion. The new information exchanged between nodes is assimilated locally to generate a global terrain height estimate on each sensor node. The distributed terrain height algorithm is demonstrated in a laboratory test environment using data from three sensor nodes over a 60 second data collection. The distributed terrain height algorithm is shown to perform equivalent to the centralized case even in the presence of communication failures.