Learning on the job: Smoothing for Simultaneous Localization and Tracking in sensor networks

Fundamental to the problem of moving target tracking is the estimation of its state with respect to the sensing device(s). However, in sensor networks, often characterized by random ad hoc deployment possibly in inaccessible or hostile environment, the locations of the sensing devices are known only to a crude approximation. We propose ConSLAT, a smoothing algorithm for Simultaneous Localization and Tracking that uses the well-known RANSAC (Random Sample Consensus) algorithm for approximation of the posterior densities. Smoothing ensures faster learning of node positions in addition to eliminating clutter. ConSLAT is completely distributed and extremely lightweight, and makes minimal assumptions about the resource availability. It requires no specific target movement patterns, and can work in the presence of multiple closely moving targets.