Title: Distributed Extended Object Tracking Information Filter Over Sensor Networks

Abstract: Motivated by the two common limitations on current distributed extended object tracking systems, i.e., the extent is modeled as an ellipse, and each sensor node in network needs to detect the object, this paper considers a multiplicative error model (MEM) to design a distributed information filter (IF) over a realistic network. The MEM reduces the extent of perpendicular axis-symmetric shapes into a 3-D vector, which results in MEM being a nonlinear and state-coupled model with multiplicative noise. To meet the requirement in IF that the state-space model is a linear model with additive noise, we first derive two separate pseudo-linearized models by using the first-order Taylor series expansion. The separation is merely in form, and the cross-correlation between two estimated states is preserved as parameters in each other's model. Thus, the joint estimation is transferred into an iterative operation of two linear filters. Second, we propose a centralized information filter by using the two models, in which the multiple measurements are converted into a summation form of innovation parts. Third, under a sensor network where the communication nodes cannot detect the object, we present two distributed information filters based on the consensus on information (CI) and consensus on measurement (CM) schemes, respectively. Finally, the performance of two distributed filters in terms of accuracy, convergence, and consistency is evaluated in Monte Carlo simulations.