Distributed Information-Weighted Consensus Filter for Extended Object Tracking With Nonlinear Measurements

This article deals with the distributed extended object tracking with nonlinear noisy measurements. Therein, we use the orientation and semiaxes as individual parameters to model the spatial extent. To alleviate nonlinearity in the measurement function, we utilize the moment-matched approach in the linear minimum mean-square error sense to statistically linearize nonlinear measurements. In this setting, the required coefficients are computed by the sampling-based method. Taking the resulting measurements as a basis, two individual linear formulas with only additive noise are fed to the information filter (IF) framework. In a consensus-based IF that exchanges the local measurements with limited iterations, the estimates are inaccurate when the weighted priors are not incorporated. Motivated by this fact, we define an integrated information-weighted consensus rule including two steps, first toward the measurement-to-measurement and then toward the global weighted fusion on the priors. This leads us to propose a distributed generalized consensus on measurement (GCM) filter to achieve an agreement on both the kinematics and extent parameters. The estimation error of the GCM filter is proven to be exponentially bounded in the mean square. Results with two types of scenarios are presented with high estimation accuracy over alternative distributed approaches.