Resilient Distributed Target Tracking Over Sensor Networks Against Misbehaving Nodes

This paper is concerned with the distributed target tracking for a moving target of discrete time-varying nonlinear dynamics over a wireless sensor network. A number of spatially distributed sensors are deployed to measure the state of the target, calculate local state predictions as well as local state estimates, and further exchange local information with their underlying neighboring sensors. Due to adversarial attacks, a subset of sensors are deliberately manipulated and thus misbehaving. Accordingly, information exchanges from the misbehaving sensors to their neighbors become antagonistic rather than cooperative as in normal operation. First, a novel distributed target tracking scheme in terms of local state predictors and state estimators is developed for each sensor over a partially misbehaving sensor network. Second, criteria for designing the desired distributed target tracking scheme and the time-varying adjacency matrix are derived such that two ellipsoidal prediction and estimation sets can be recursively computed. It is analytically proved that these two sets guarantee the containment of the true target state at every instant of time regardless of misbehaving sensors as well as unknown-but-bounded process and measurement noises. Third, based on the proposed design criteria, optimization methods are put forward to minimize the calculated ellipsoids. Finally, an application to vehicle tracking is given to show the effectiveness of the results.